Modular Invariant Theory

Encyclopaedia of Mathematical Sciences Volume 139 Invariant Theory and Algebraic Transformation Groups VIII Subseries Editors: Revaz V. Gamkrelidze Vladimir L. Popov

H.E.A. Eddy Campbell r David L. Wehlau

Modular Invariant Theory

H.E.A. Eddy Campbell University of New Brunswick Sir Howard Douglas Hall Dept. Mathematics Bailey Drive 3 Fredericton, New Brunswick Canada E3B 5A3 [email protected]

David L. Wehlau Royal Military College of Canada Dept. Mathematics & Computer Science Kingston, Ontario Canada K7K 7B4 [email protected]

Founding editor of the Encyclopaedia of Mathematical Sciences: Revaz V. Gamkrelidze

ISSN 0938-0396 ISBN 978-3-642-17403-2 e-ISBN 978-3-642-17404-9 DOI 10.1007/978-3-642-17404-9 Springer Heidelberg Dordrecht London New York Mathematics Subject Classification (2010): 13A50 © Springer-Verlag Berlin Heidelberg 2011 This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Cover design: deblik Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)

For Diane Mary Brennan, Ian Alexander Brennan, Colin Cameron Brennan, Graham Harold James Brennan and Maggie Orion Cameron. For Charlene Lynn Janeway and Megan Melinda Jane Wehlau.

Preface

At the time we write this book there are several excellent references available which discuss various aspects of modular invariant theory from various points of view: Benson [6]; Derksen and Kemper [26]; Neusel [85]; Neusel and Smith [86]; and Smith [103]. In this book, we concentrate our attention on the modular invariant theory of finite groups. We have included various techniques for determining the structure of and generators for modular rings of invariants, while attempting to avoid too much overlap with the existing literature. An important goal has been to illustrate many topics with detailed examples. We have contrasted the differences between the modular and non-modular cases, and provided instances of our guiding philosophies and analogies. We have included a quick survey of the elements of algebraic geometry and commutative algebra as they apply to invariant theory. Readers who are familiar with these topics may safely skip this chapter. We wish to thank our principal collaborators over the years with whom we have had so much pleasure exploring this fascinating subject: Ian Hughes, Gregor Kemper, R. James Shank, John Harris as well as our students and friends, Jianjun Chuai, Greg Smith, Mike Roth, Brandon Fodden, Emilie Dufresne, Asia Matthews and Chester Weatherby. In particular we thank John Harris, R. James Shank, Jianjun Chuai, Mike Roth, Emilie Dufresne, Asia Matthews, Chester Weatherby and Tristram Bogart for reading draft chapters and pointing out errors and suggesting improvements. We also thank Marie-Jos´e Bertin for clarifying the history of her own work to us. Finally, our thanks go to the anonymous referees for many helpful and constructive remarks.

Fredericton, New Brunswick Kingston, Ontario August 2010

H. E. A. Eddy Campbell David L. Wehlau

Contents

1

2

First Steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Groups Acting on Vector Spaces and Coordinate Rings . . . . . . . 1.1.1 V Versus V ∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Constructing Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 On Structures and Fundamental Questions . . . . . . . . . . . . . . . . . 1.4 Bounds for Generating Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 On the Structure of K[V ]G : The Non-modular Case . . . . . . . . . . 1.6 Structure of K[V ]G : Modular Case . . . . . . . . . . . . . . . . . . . . . . . . . 1.7 Invariant Fraction Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.8 Vector Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.9 Polarization and Restitution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.10 The Role of the Cyclic Group Cp in Characteristic p . . . . . . . . . 1.11 Cp Represented on a 2 Dimensional Vector Space in Characteristic p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.12 A Further Example: Cp Represented on 2 V2 in Characteristic p 1.13 The Vector Invariants of V2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 2 4 6 7 7 8 9 10 11 11 16 17 20 23

Elements of Algebraic Geometry and Commutative Algebra 2.1 The Zariski Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 The Topological Space Spec(S) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Noetherian Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Localization and Fields of Fractions . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Integral Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6 Homogeneous Systems of Parameters . . . . . . . . . . . . . . . . . . . . . . 2.7 Regular Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.8 Cohen-Macaulay Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.9 The Hilbert Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.10 Graded Nakayama Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.11 Hilbert Syzygy Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25 25 27 27 29 29 30 31 32 34 35 36

X

3

4

Contents

Applications of Commutative Algebra to Invariant Theory . 3.1 Homogeneous Systems of Parameters . . . . . . . . . . . . . . . . . . . . . . 3.2 Symmetric Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 The Dickson Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Upper Triangular Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Noether’s Bound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 Representations of Modular Groups and Noether’s Bound . . . . 3.7 Molien’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.7.1 The Hilbert Series of the Regular Representation of the Klein Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.7.2 The Hilbert Series of the Regular Representation of C4 . 3.8 Rings of Invariants of p-Groups Are Unique Factorization Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.9 When the Fixed Point Subspace Is Large . . . . . . . . . . . . . . . . . . .

39 40 44 45 46 46 48 50 51 53 54 55

Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 The Cyclic Group of Order 2, the Regular Representation . . . . 4.2 A Diagonal Representation of C2 . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Fraction Fields of Invariants of p-Groups . . . . . . . . . . . . . . . . . . . 4.4 The Alternating Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Invariants of Permutation Groups . . . . . . . . . . . . . . . . . . . . . . . . . 4.6 G¨ obel’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.7 The Ring of Invariants of the Regular Representation of the Klein Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.8 The Ring of Invariants of the Regular Representation of C4 . . . 4.9 A 2 Dimensional Representation of C3 , p = 2 . . . . . . . . . . . . . . . 4.10 The Three Dimensional Modular Representation of Cp . . . . . . . 4.10.1 Prior Knowledge of the Hilbert Series . . . . . . . . . . . . . . . . 4.10.2 Without the Use of the Hilbert Series . . . . . . . . . . . . . . . .

59 61 62 62 64 65 66 69 72 75 75 76 78

5

Monomial Orderings and SAGBI Bases . . . . . . . . . . . . . . . . . . . . 5.1 SAGBI Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Symmetric Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Finite SAGBI Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 SAGBI Bases for Permutation Representations . . . . . . . . . . . . . .

83 85 89 91 93

6

Block Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 6.1 A Block Basis for the Symmetric Group . . . . . . . . . . . . . . . . . . . . 101 6.2 Block Bases for p-Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

7

The Cyclic Group Cp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 7.1 Representations of Cp in Characteristic p . . . . . . . . . . . . . . . . . . . 105 7.2 The Cp -Module Structure of F[Vn ] . . . . . . . . . . . . . . . . . . . . . . . . . 110 7.2.1 Sharps and Flats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 7.3 The Cp -Module Structure of F[V ] . . . . . . . . . . . . . . . . . . . . . . . . . . 113

Contents

XI

7.4 The First Fundamental Theorem for V2 . . . . . . . . . . . . . . . . . . . . 115 7.4.1 Dyck Paths and Multi-Linear Invariants . . . . . . . . . . . . . . 117 7.4.2 Proof of Lemma 7.4.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 7.5 Integral Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 7.6 Invariant Fraction Fields and Localized Invariants . . . . . . . . . . . 130 7.7 Noether Number for Cp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 7.8 Hilbert Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 8

Polynomial Invariant Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 8.1 Stong’s Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 8.2 A Counterexample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 8.3 Irreducible Modular Reflection Groups . . . . . . . . . . . . . . . . . . . . . 149 8.3.1 Reflection Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 8.3.2 Groups Generated by Homologies of Order Greater than 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 8.3.3 Groups Generated by Transvections . . . . . . . . . . . . . . . . . 151

9

The Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 9.1 The Transfer for Nakajima Groups . . . . . . . . . . . . . . . . . . . . . . . . 164 9.2 Cohen-Macaulay Invariant Rings of p-Groups . . . . . . . . . . . . . . . 170 9.3 Differents in Modular Invariant Theory . . . . . . . . . . . . . . . . . . . . . 173 9.3.1 Construction of the Various Different Ideals . . . . . . . . . . . 174

10 Invariant Rings via Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 11 Rings of Invariants which are Hypersurfaces . . . . . . . . . . . . . . . 185 12 Separating Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 12.1 Relation Between K[V ]G and Separating Subalgebras . . . . . . . . 195 12.2 Polynomial Separating Algebras and Serre’s Theorem . . . . . . . . 198 12.3 Polarization and Separating Invariants . . . . . . . . . . . . . . . . . . . . . 201 13 Using SAGBI Bases to Compute Rings of Invariants . . . . . . . 205 14 Ladders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 14.1 Group Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 14.2 Cohomology and Invariant Theory . . . . . . . . . . . . . . . . . . . . . . . . . 214 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

Index of notations N Z Q R C F F× Fp K K ⊗ V GL(V ) GLn (K) SL(V ) SLn (K) G σ, τ σ(v), σ · v 2 VG 2 2 GX Gv 2 V∗ 3 xI F[V ] 3 F[V ]G 4 6 TrG (f ) TrG 6 H (f ) 6 N(f )G NG (f ) 7 H Gf 7 β(G, V ) 8 |G| 8 Vσ K(V )G 10 Quot(K[V ]G ) 10 mV 11 Pol 11 R 11 degy (f ) 18 Δσ 22 V(T ) 26 I(X) 26

the natural numbers, including 0. the integers. the rational numbers. the real numbers. the complex numbers. a field of characteristic p = 0. F \ {0}. the prime field of order p. a field of arbitrary characteristic. the algebraic closure of K. ⊗K or ⊗F . a vector space over F or K, most often of dimension n. the general linear group of V . the linear automorphisms of Kn . the special linear group of V . the linear automorphisms of Kn of determinant 1. a group, often a subgroup of GL(V ). group elements. the image v ∈ V under the action of σ. the subspace of V fixed (point-wise) by the action of G. the stabilizer or isotropy subgroup of X ⊂ V . the stabilizer or isotropy subgroup of v ∈ V . the hom-dual homK (V, K). the monomial xi11 · · · xinn for I = (i1 , . . . , in }. the coordinate ring of V . the invariant ring. the trace of f ∈ F[V ]. the trace of f ∈ F[V ] relative to a subgroup H ⊂ G. the norm of f . the norm of f relative to a subgroup H ⊂ G. the orbit of f under G, i.e., Gf := {σ(f ) | σ ∈ G}. the Noether degree bound for the invariant ring. the order of G. the elements of V fixed by σ. the field of fractions of K[V ]G . the field of fractions of K[V ]G . the vector space V ⊕m . polarization. restitution. the degree of f in the variable y. the twisted derivation σ − 1. the zero-set of the polynomials in T . the ideal of the variety or set X.

Index of notations

√

I Spec(S) MaxSpec(S) R Rf H(R, λ) V //G (V //G)good LM(f ) LT(f ) LT(R) K[V ]G Ga (C) RLn (F) GUn (F) RUn (F) GOn (F) SOn (F) Ωn± (F)

XIII

26 the radical of the ideal I. 27 the space of all prime ideals of S. 27 the space of all maximal prime ideals of S. a graded commutative ring with 1. 29 the localization of R at f . 34 the Hilbert series of R. 42 the quotient variety. 56 the “good” locus. 83 the lead monomial of f (with respect to some ordering). 83 the lead term of f . 85 the lead term algebra of R. 99 the ring of coinvariants. 125 the additive group of C. 150 the subgroup of GLn (F) generated by its reflections. 150 the unitary group of dimension n over F. 150 the subgroup of GUn (F) generated by its reflections. 150 the general orthogonal group over F. 150 the special orthogonal group over F. 150 the commutator subgroup of GOn (F).

1 First Steps

Invariant theory seeks to determine whether a (mathematical) object can be obtained from some other object by the action of some group. One way to answer this question is to find some functions that map from the class of objects to some field (or more generally some ring). Invariants are functions which take the same value on any two objects which are related by an element of the group. Thus if we can find any invariant which takes different values on two objects, then these two objects cannot be related by an element of the group. Ideally, we hope to find enough invariants to separate all objects which are not related by any group element. This means we want to find a (finite) set of invariants f1 , f2 , . . . , fr with the property that if two objects are not related by the group action then at least one of these r invariants takes different values on the two objects in question. For example, suppose we wish to determine whether two triangles are congruent, that is, whether one can be obtained from the other by translation, rotation, reflection or a combination of these operations. One useful invariant is the area function: two triangles with different areas cannot be congruent. On the other hand, the (unordered) set of three functions which give the lengths of the three sides are sufficient: two different triangles having sides of the same lengths must be congruent. For us, the mathematical objects are elements of some vector space with a group action and the invariants will be those regular functions on the vector space that are constant on each of the group orbits. We begin with some basic material on the action of groups on vector spaces and their coordinate rings, followed by a simple illustrative example. There are excellent references available: Benson [6], Derksen and Kemper [26], Neusel [85], Neusel and Smith [86] and Smith [103]. We also note that many advances in modular invariant theory have been made due to the programming language MAGMA [10], especially the invariant theory packages developed by Gregor Kemper.

H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 1, © Springer-Verlag Berlin Heidelberg 2011

1

2

1 First Steps

1.1 Groups Acting on Vector Spaces and Coordinate Rings We begin with a finite dimensional representation ρ of a group G over a field K, i.e., a group homomorphism ρ : G → GL(V ) where V is a finite dimensional vector space over K. In this book we will always denote the characteristic of the field (which may be 0) by p. For us, the group G is always assumed to be finite of order denoted |G|. The representation of G is said to be a modular representation if p divides |ρ(G)| and a non-modular one if not. Many questions which are well understood in the non-modular case are much less well understood in the modular case. One of the main reasons for this is that Maschke’s Theorem fails to hold for modular representations. That is, modular representations may not be completely reducible and usually are not. Researchers have substituted techniques from algebraic geometry, commutative algebra, and group cohomology (including Steenrod operations) in an effort to make up for this deficiency. The main technique used in this book is the fact from representation theory that the cyclic group of order p has only finitely many indecomposable inequivalent representations in characteristic p. The representation defines a left action of the group G on V . Given σ ∈ G and v ∈ V we write σ(v) for the vector ρ(σ)(v), the result of applying ρ(σ) to v. Very often, the representation is fixed throughout an example and we make little reference to it. We denote the set of vectors fixed (pointwise) by the group G by V G = {v ∈ V | σ(v) = v, for all σ ∈ G}, and for a subset X of V , we denote by GX = {σ ∈ G | σ(v) = v, for all v ∈ X} the isotropy or stabilizer subgroup of X. Usually, if X = {v} is a singleton set we will write Gv to denote GX . Now consider V ∗ , the vector space dual to V . This is the set, HomK (V, K), of linear functionals from V to K. Recall that x : V → K is said to be a linear functional if x(av + bw) = ax(v) + bx(w) for all v, w ∈ V , and all a, b ∈ K. Of course, we have dimK (V ∗ ) = dimK (V ). The action of G on V determined by ρ naturally induces a left action of G on V ∗ as follows. Let x ∈ V ∗ be any linear functional on V and let σ ∈ G. Then σ(x) should be another linear functional on V . This new linear functional is defined by (σ(x))(v) := x(σ −1 (v)). In this definition we use σ −1 instead of σ in order to obtain a left action (and not a right action) of G on V ∗ . This new representation of G is often referred to as the dual representation.

1.1 Groups Acting on Vector Spaces and Coordinate Rings

3

Lemma 1.1.1. Suppose we have a fixed representation ρ : G → GL(V ) and consider also ρ∗ : G → GL(V ∗ ). In general, for σ ∈ G the matrix representing ρ(σ) ∈ GL(V ) with respect to a fixed basis is the transpose inverse of the matrix representing ρ∗ (σ) with respect to the dual basis.   Associated to the vector space V is its coordinate ring also called its ring of regular functions. This ring, denoted K[V ], is a major object of study in algebraic geometry. We may define K[V ] in a number of equivalent ways. Here is a very concrete definition of the coordinate ring of V . Let {x1 , x2 , . . . , xn } be a fixed basis of V ∗ . Then K[V ] is the polynomial ring in n variables: K[V ] = K[x1 , x2 , . . . , xn ]. This is a useful description of K[V ]. For an exponent sequence I = (i1 , . . . in ) consisting of non-negative integers, we define the monomial xI = xi11 · · · xinn , I We say that xI has degree i1 + · · · + in and we denote the degree  of xIj by I deg(x ) or even deg(I). As usual, we say that a polynomial f = aj x for ai ∈ K is homogeneous of degree d if each of its monomials, xIj , is of degree d. We observe that K[V ] is naturally graded by degree: we may write

K[V ] = ⊕d≥0 K[V ]d where K[V ]d denotes the subspace of homogeneous polynomials of degree d (including the zero polynomial). We also observe that K[V ] is a graded algebra. This just means that each K[V ]d is a subspace and that if f ∈ K[V ]d and f  ∈ K[V ]d then f f  ∈ K[V ]d+d . If V is a direct sum, V = U ⊕ W then we have a finer grading indexed by N2 on K[V ] induced by the isomorphism K[V ] ∼ = K[U ] ⊗K K[W ] given by K[V ](d,d ) = K[U ]d ⊗K K[W ]d . More generally, if V = W1 ⊕ W2 ⊕ · · · ⊕ Wm then K[V ] has a Nm -grading given by K[V ](d1 ,d2 ,...,dm ) = K[W1 ]d1 ⊗ K[W2 ]d2 ⊗ · · · ⊗ K[Wm ]dm . Thus if f ∈ K[V ](d1 ,d2 ,...,dm ) , then f (t1 v1 , t2 v2 , . . . , tm vm ) = td11 td22 · · · tdmm f (v1 , v2 , . . . , vm ) for all t1 , t2 , . . . , tm ∈ K. We say that elements of K[V ](d1 ,d2 ,...,dm ) are multihomogeneous. If each Wi is G-stable, then the G-action will stabilize each Nm -graded summand K[V ](d1 ,d2 ,...,dm ) of K[V ].

4

1 First Steps

From an abstract point of view, if K is infinite we may define K[V ] as a ring of functions: K[V ] := {f : V → K | f is a regular function on V }. A function f is regular on V if f may be written as a polynomial in some (and hence every) basis of linear functionals on V . We note that in order to view K[V ] as a ring of functions on V we require that K be infinite. If K is finite, for example, if K = Fp , the field with p elements, then the two different polynomials x1 and xp1 in K[V ] determine the same function on V . Let K denote an algebraic closure of K and let V = K⊗V . The inclusion K ⊂ K induces an inclusion V ⊂ V . Thus K[V ] ⊆ K[V ] and two elements of K[V ] are equal if and only if they determine the same function on V. We may also define K[V ] as the symmetric algebra on V ∗ : K[V ] = S • (V ∗ ). The action of G on V given by ρ also naturally induces an action of G on K[V ]. We may describe this action in two ways according to the description we use for K[V ]. In terms of polynomials we merely extend the action of G on V ∗ additively and multiplicatively. That is, let σ ∈ G and f, f  ∈ K[V ]. Thus σ(f + f  ) = σ(f ) + σ(f  ) and σ(f f  ) = (σ(f ))(σ(f  )). Equivalently, if we regard the elements of K[V ] as functions on V we may define this action of G on K[V ] via (σ(f ))(v) := f (σ −1 (v)). The main object of study in invariant theory is the collection of polynomial functions on V left fixed by all of G. This collection of functions forms a ring, denoted K[V ]G : K[V ]G := {f ∈ K[V ] | σ(f ) = f for all σ ∈ G}. We observe that if a polynomial f is fixed by both σ and τ ∈ G, then f is also fixed by στ . We may conclude, therefore, that if f is invariant with respect to every element of some set of generators for G, then f ∈ K[V ]G . 1.1.1 V Versus V ∗ A common question that arises is why we insist upon considering the action of G upon V ∗ and K[V ] rather than on the symmetric algebra on V , S • (V ). In order to answer this question, consider the following example. Example 1.1.2. Let G = Cp × Cp , the elementary Abelian p-group of order p2 . We consider a three dimensional representation of G given by ⎫ ⎧⎛ ⎞ ⎬ ⎨ 100

G = ⎝a 1 0⎠ a, b ∈ Fp ⊂ GL(V ). ⎭ ⎩ b01

1.1 Groups Acting on Vector Spaces and Coordinate Rings

5

Here G is generated by the two elements given by taking (a, b) = (1, 0) and (a, b) = (0, 1). These two elements are easily seen to be of order p and to commute. Thus G is indeed isomorphic to Cp × Cp . We examine the geometry of the action of G on V by considering the orbits under this action. Let v = (v1 , v2 , v3 ) ∈ V ∼ = F3p . If v1 = 0 then we see that G · v = {(v1 , v2 + av1 , v3 + bv1 ) | a, b ∈ Fp } consists of p2 points. Conversely, if v1 = 0 then the orbit of v consists of the single point v. If σ ∈ G is represented by a matrix A in GL(V ) with respect to the standard basis, then the matrix of σ in GL(V ∗ ) with respect to the dual basis is given by (AT )−1 . Thus working with the basis of V ∗ dual to the standard basis of V , we see that ⎧⎛ ⎫ ⎞ ⎨ 1 −a −b

⎬

G = ⎝0 1 0 ⎠ a, b ∈ Fp ⊂ GL(V ∗ ). ⎩ ⎭ 0 0 1 Let {e1 , e2 , e3 } be the standard basis of V and let {x1 , x2 , x3 } denote the dual basis of V ∗ . The geometry of G acting on V is reflected in the invariant x2 and functions in Fp [V ]G = S • (V ∗ )G = F[x1 , n2 , n3 ] where n2 = xp2 − xp−1 1 x3 . If we consider a point v with 0 = v1 ∈ Fp then the two n3 = xp3 − xp−1 1 functions n2 and n3 are both constant on these orbits. Moreover, it is not too 3 difficult to see that, if v ∈ V ∼ = Fp with x1 (v) = x1 (v ), n2 (v) = n2 (v ) and   n3 (v) = n3 (v ), then v ∈ Gv. Using S • (V )G instead, we would have found S • (V )G = Fp [f1 , e2 , e3 ] where f1 is a cubic expression beginning f1 = e31 + . . . . In particular, these do not correspond to functions which are constant on the orbits of G. This example shows why we are interested in both the matrix representation of G on V and also on V ∗ . Examining the former allows us to see the geometry of the group action. Examining the latter allows us to understand which polynomials are invariants. Rather than writing out both matrices for a group element σ, we will often compromise by writing out the matrix A−1 of σ −1 in GL(V ). This shows us directly how σ −1 is acting on V and allows us to study the orbits in V . The transpose of this matrix shows how σ acts on V ∗ and thus we may understand the action of σ on V ∗ by considering the rows of A−1 and the action of A−1 on row vectors by right multiplication. A dramatic illustration of the difference between the group actions on V and V ∗ is provided by the following subgroup of GL(V ) where V is a seven dimensional vector space over Fp , the field of order p. We define ⎛ ⎞ 1000000 ⎜ 0 1 0 0 0 0 0⎟ ⎜ ⎟ ⎜ 0 0 1 0 0 0 0⎟ ⎜ ⎟ ⎟ σ(a, b, c, d) = ⎜ ⎜a 0 0 1 0 0 0⎟ ⎜ 0 b 0 0 1 0 0⎟ ⎜ ⎟ ⎝0 0 c 0 0 1 0⎠ ddd0001

6

1 First Steps

and we take G = {σ(a, b, c, d) | a, b, c, d ∈ Fp } ⊂ GL(V ). We will show in Example 8.0.8 that F[V ]G is a polynomial ring. On the other hand, consider the group H ⊂ GL(V ) consisting of the transposes of the elements of G acting on V , that is, the group of matrices ⎛ ⎞ 100a00d ⎜0 1 0 0 b 0 d⎟ ⎜ ⎟ ⎜0 0 1 0 0 c d⎟ ⎜ ⎟ ⎟ τ (a, b, c, d) = ⎜ ⎜0 0 0 1 0 0 0⎟ . ⎜0 0 0 0 1 0 0⎟ ⎜ ⎟ ⎝0 0 0 0 0 1 0⎠ 0000001 We will show in Example 8.0.9 that F[V ]H is not Cohen-Macaulay. Note that both groups are generated by reflections (reflections are defined below in §1.5).The definitions of polynomial rings and Cohen-Macaulay rings may be found in §2.3 and §2.8 respectively.

1.2 Constructing Invariants One general method to construct invariants of finite groups is as follows. Let f ∈ K[V ]. Then the transfer or trace of f is defined as  Tr(f ) = TrG (f ) := σ(f ). g∈G

Similarly, the norm of f is defined by N(f ) = NG (f ) :=



σ(f ).

g∈G

We also have occasion to use relative versions of these constructions. Suppose H is a subgroup of G and we have a polynomial f which is H-invariant. Then we choose a fixed set of left coset representatives G/H := {σ1 , σ2 , . . . , σr } and define r  (f ) := σ (f ) TrG H =1

and NG H (f ) :=

r 

σ (f ).

=1 G It is easy to see that for f ∈ K[V ]H the elements TrG H (f ) and NH (f ) are independent of the choice of σ1 , σ2 , . . . , σr . However, for general f this is not true. For this reason, it is often useful to take H to be the isotropy subgroup

1.4 Bounds for Generating Sets

7

Gf . Of course, for any finite group G, subgroup H, and f ∈ K[V ]H we have G G G that TrG H (f ) and NH (f ) are both G-invariants. Note that Tr (f ) = Tr{e} (f ) G G and N (f ) = N{e} (f ). Still more generally, consider an element f ∈ K[V ]. We define the G-orbit of f to be {σ(f ) | σ ∈ G}, denoted Gf . A slightly different way to describe the orbit of f is to use the isotropy subgroup Gf of f . We have Gf = {σ(f ) | σ ∈ G/Gf } where G/Gf denotes a set of (left) coset representatives of Gf in G. Suppose, then, that |Gf | = m. From here, we can form the polynomial Sf (λ) =

 h∈Gf

(λ − h) =

m  (−1)i si λm−i , i=0

where si ∈ K[V ]G . The coefficients are elementary symmetric functions in the elements of Gf . That is, if we write Gf = {f1 , . . . , fm }, then s1 (f ) = f1 + f2 + · · · fm , s2 (f ) = f1 f2 + f1 f3 + · · · + fm−1 fm , .. . sm (f ) = f1 f2 · · · fm . For any finite group G, and f ∈ K[V ] we have that TrG (f ) = |Gf |s1 (f ), NG (f ) = sm (f )|Gf | .

1.3 On Structures and Fundamental Questions The problems we will consider fall roughly into two classes: 1. Find generators for K[V ]G . Failing that, find an upper bound for the largest degree of an element of a homogenous minimal generating set. 2. Determine the structure of K[V ]G . For example, determine those groups G for which the ring of invariants K[V ]G is a polynomial algebra, a hypersurface, or a Cohen-Macaulay ring. Both questions are interesting for either specific groups, or for classes of groups. In general, much more is known in the non-modular case than in the modular case.

1.4 Bounds for Generating Sets Emmy Noether proved (see Theorem 3.1.2) that the ring of invariants of a representation V of a finite group acting is always generated as an algebra by a finite collection of homogeneous invariants f1 , f2 , . . . ft . Using the

8

1 First Steps

graded Nakayama lemma (Lemma 2.10.1) we see that the number β(V, G) := max{deg(fi ) | 1 ≤ i ≤ t} is independent of the choice of generators provided t is minimal. This number β(V, G) is called the Noether number for V . Noether showed that generators of degree at most |G| are required when p = 0. For non-modular groups with p > |G|, this theorem is still true. Richman and others have shown Noether’s original bound, β(V, G) ≤ |G|, applies if G is solvable. Smith [103][pg 175], Fleischmann [39], and others have shown that for non-modular groups, K[V ]G is generated in degrees at most dimK (V )(|G| − 1). For an overview of this topic, see Wehlau’s paper [111]. Here we need dimK (V ) > 1 and |G| > 1. There was a conjecture that non-modular groups have rings of invariants that are generated in degrees less than or equal |G|. The difference between the known bound and this conjectural bound was known as the problem of Noether’s gap: is there a non-modular group in the gap or not? In 1999, Fleischmann gave a beautiful and clever variation of Noether’s original argument that showed the conjecture was true (see [39]). Independently, Fogarty [42] proved the same result. Below we give a simplified version of Fogarty’s proof, due to Benson, see Theorem 3.5.1. It is proved by Campbell, Geramita, Hughes, Shank and Wehlau in [17] that if K[V ]G is a hypersurface, then this ring is generated in degrees less than or equal to |G|. More generally, Broer [12] has shown that if K[V ]G is CohenMacaulay, then this ring of invariants is generated by elements of degree at most dimK (V )(|G| − 1). G. Kemper has made the conjecture that Noether’s degree bound, |G|, applies whenever K[V ]G is Cohen-Macaulay. Symonds [106], using work of Karagueuzian and Symonds [62] has proved that Theorem 1.4.1. If K is finite and G is a non-trivial finite group acting on V with dimK (V ) > 1, then K[V ]G is generated in degrees less than or equal to dimK (V )(|G| − 1). A synopsis of this work is given in §3.6.

1.5 On the Structure of K[V ]G: The Non-modular Case The invariant theory of finite groups is much better understood in the nonmodular case. For example, in this situation, a complete characterization of those representations for which K[V ]G is polynomial is known. To state this characterization we need the following definition. Definition 1.5.1. Let V be a representation of G defined over a field K. Then σ ∈ G is a reflection if dim V σ = dim V − 1. Over a field of characteristic p, a reflection of order p is called a transvection. Classically, a (real) reflection was defined as an element σ with single nontrivial eigenvalue −1, and a (complex) reflection as an element with single

1.6 Structure of K[V ]G : Modular Case

9

non-trivial eigenvalue a (complex) root of unity. What we have defined as a “reflection” was originally called a “pseudo-reflection”. This older terminology is still used by some authors. The following famous and beautiful theorem follows from the work of Coxeter [24], Shephard and Todd [101], Chevalley [22], and Serre [95]. To prove one direction, that groups generated by reflections over C have polynomial invariant rings, Shephard and Todd classified all such representations and showed that in each case the ring of invariants is polynomial. Unaware of their work, Chevalley [22] proved in 1955 that for representations over R generated by reflections of order 2 the ring of invariants is always polynomial. Chevalley’s proof is truly beautiful, short and does not rely on any classification. Serre who was familiar with the work of Shephard and Todd observed that Chevalley’s proof works for all groups generated by reflections over C not just reflections of order 2. He also proved a partial converse valid over any field, see below. We describe a new proof by Dufresne of this result, see Section 12.2. Theorem 1.5.2. Let G be a finite group with |G| invertible in the field K. Then K[V ]G is a polynomial algebra if and only if the action of G on V is generated by reflections. Theorem 1.5.3. Let G be a finite group represented over F. If K[V ]G is a polynomial algebra then the action of G on V is generated by reflections. Aside from examples and special cases (see for example Nakajima’s Theorem 8.0.7), the characterization of representations of finite groups with polynomial rings of invariants remains one of the most important open problems in modular invariant theory. There are other wonderful theorems concerning characterizations of hypersurfaces (Nakajima), Gorenstein rings (Watanabe), or Cohen-Macaulay rings (Hochster and Eagon) in the non-modular case. In the modular case, we note the theorem of Kemper [65]: a bi-reflection is an element σ ∈ G with Im(σ − 1 : V → V ) of dimension less than or equal to 2. Theorem 1.5.4. Let G be a finite group with |G| represented over the field F of characteristic p > 0 with p | |G|. If K[V ]G is Cohen-Macaulay then if the action of G on V is generated by bi-reflections. This topic is explored in more depth in §9.2. It remains an open problem to characterize those modular bi-reflection groups whose rings of invariants are Cohen-Macaulay.

1.6 Structure of K[V ]G: Modular Case J.P. Serre proved one direction of Theorem 1.5.2 holds in the modular case. He showed that whenever K[V ]G is a polynomial ring, the action of G on V must

10

1 First Steps

be generated by reflections. Examples of reflection groups whose invariant rings are not polynomial are known. See for example, §8.2. Nakajima has characterized those p-groups with polynomial rings of invariants when K = Fp is the prime field of order p. Roughly speaking, he shows that such groups resemble the ring of invariants of the full Upper Triangular group. He gave examples of elementary Abelian reflection p-groups with non-Cohen-Macaulay invariant rings, a somewhat simpler example is the example mentioned above at the end of §1.1. Nakajima’s characterization fails over larger fields, as shown by an example due to Stong, see §8.1. Kemper and Malle have examined the class of irreducible representations of modular reflection groups and determined which have polynomial rings of invariants. Unfortunately, irreducible representations are few and far between. We summarize their work in §8.3. Much work remains to be done on characterizing groups with polynomial rings of invariants.

1.7 Invariant Fraction Fields It will be useful on occasion to study the fraction fields denoted Quot(K[V ]) or K(V ) and K(V )G of the domains K[V ] and K[V ]G , respectively; in some situations we encounter, it is useful to recall the results of Galois Theory. It is not difficult to see that (K(V ))G = Quot(K[V ]G ). For, given an invariant fraction ff ∈ K(V )G , we may write   f σ=1 σ(f  ) f σ=1 σ(f  ) f = =  f f σ=1 σ(f  ) NG (f  ) and note that the denominator of the right hand side is invariant. Since the fraction itself is also invariant, the numerator of the right hand side is invariant as claimed. Then we have the diagram K[V ]G

⊂

∩

? K(V )G

- K[V ] ∩

⊂

? - K(V )

and we see that the bottom row of this diagram tells us that K(V ) is a Galois extension of K(V )G , that is, there exist q = |G|-many rational functions ai = ffi such that {a1 , . . . aq } is a basis for K(V ) as a vector space over K(V )G . i Furthermore, the induced G-action on K(V ) = ⊕qi=1 K(V )G ai

1.9 Polarization and Restitution

11

is the regular representation of G. It is a famous question of Noether’s whether or not K(V )G is purely transcendental; this is the question of whether or not there are n = dim(V ) elements ai ∈ K(V )G such that K(V )G = K(a1 , . . . , an ). The answer to this question is negative in general. However, if p > 0 and G is a p-group, then K(V )G is purely transcendental (see [81]). We will revisit this question in section §7.6.

1.8 Vector Invariants Consider the coordinate ring of m V = ⊕m V with the diagonal action of G. The ring K[m V ]G is called a ring of vector invariants of G. Rings of vector invariants provide an important class of examples and counterexamples. In [19], Campbell and Hughes give generators, as conjectured by Richman [92], for Fp [m V2 ]Cp where Cp denotes the cyclic group of order p, and V2 denotes its 2 dimensional indecomposable representation. An easy corollary is the fact, first observed by Richman, that this invariant ring requires a generator of degree m(p − 1). Therefore, Noether’s degree bound, |G|, does not hold for p-groups. Kemper has proved that, if G is any modular group, then F[m V ]G is not Cohen-Macaulay for all sufficiently large m. In every example known, taking m = 3 is sufficiently large to obtain a non-Cohen-Macaulay ring of invariants. If G is a p-group and m ≥ 3, then F[mV ]G is not Cohen-Macaulay, see 9.2.3. This is an important corollary of the result (see 9.2.2) that if K[V ]G is Cohen-Macaulay, then G is generated by bi-reflections. Here an element σ ∈ G is called a bi-reflection if dim V σ ≥ dim V − 2. This theorem shows us how rarely we may expect to encounter Cohen-Macaulay rings as the invariants of p-groups.

1.9 Polarization and Restitution Consider the maps Δ : V → m V = V ⊕ V ⊕ · · · ⊕ V and φ : m V → V given    m copies

by Δ(v) = (v, v, . . . , v) and φ(v1 , v2 , . . . , vm ) = v1 + v2 + · · · + vm . Both of these maps are GL(V )-equivariant where GL(V ) acts diagonally on m V . These two maps naturally induce ring maps Δ∗ : F[m V ] → F[V ] and ∗ φ : F[V ] → F[m V ] given by (Δ∗ (F ))(v) = F (Δ(v)) = F (v, v, . . . , v) and (φ∗ (f ))(v1 , v2 , . . . , vm ) = f (φ(v1 , v2 , . . . , vm )) = f (v1 + v2 + · · · + vm ). Let f ∈ F[V ]d . Using the Nm -grading on F[m V ] we have  φ∗ (f ) = f(i1 ,i2 ,...,im ) i1 +i2 +···+im =d

12

1 First Steps

where each f(i1 ,i2 ,...,im ) ∈ F[m V ](i1 ,i2 ,...,im ) . These polynomials f(i1 ,i2 ,...,im ) are called partial polarizations of f and we write   Polm (f ) = f(i1 ,i2 ,...,im ) | i1 + i2 + · · · + im = d to denote the set of all such partial polarizations. In order to compute individual polarizations, we take m indeterminates λ = (λ1 , λ2 , . . . , λm ), and consider v = (v1 , v2 , . . . , vm ) where each vi represents a generic element of V . We write λv = λ1 v1 + λ2 v2 + · · · + λm vm , and then we have f (λv) = φ∗ (f )(λ1 v1 , λ2 v2 , . . . , λm vm )  λi11 λi22 · · · λimm f(i1 ,i2 ,...,im ) (v1 , v2 , . . . vm ) = i1 +i2 +...im =d

=



λI fI (v)

|I|

with |I| = i1 + i2 + · · · + im = d where fI ∈ K[m V ]I = K[V ]i1 ⊗ K[V ]i2 ⊗ · · · ⊗ K[V ]im ⊂ K[m V ]d . As a special case, we may take m = d = deg(f ) and (i1 , i2 , . . . , im ) = (1, 1, . . . , 1) to get the full polarization of f denoted P(f ) = f(1,1,...,1) = fmulti-linear ∈ K[d V ]. Lemma 1.9.1. The mapping f → f(i1 ,i2 ,...,im ) is GL(V )-equivariant. In particular, if G is any subgroup of GL(V ) and f ∈ K[V ]G , then Polm (f ) ⊂ K[m V ]G . Proof. Let σ ∈ GL(V ). We need to show (σf )I = σ(fI ). The former is defined by the equation  (σf )(λv) = λI (σf )I (v). I

But

(σf )(λv) = f (λσ −1 v) =



λI fI (σ −1 v).

I

Therefore,

(σf )I (v) = fI (σ −1 (v)) = (σfI )(v).  

Lemma 1.9.2. The full polarization P(f ) of f is a symmetric function, i.e., P(f )(τ (v)) = P(f )(v) where τ (v) = (vτ (1) , . . . , vτ (m) ) for all τ ∈ Σm .

1.9 Polarization and Restitution

13

Proof. Since τ (λv) = (λτ (1) vτ (1) + λτ (2) vτ (2) + · · · + λτ (m) vτ (m) ) we have f (λv) = f (τ (λv)) for all τ ∈ Σd .

 

The map induced by Δ is called restitution and denoted by R or by Rm . It is defined by R : K[m V ] → K[V ] and R(F )(v) = F (v, v, . . . , v).    m

lemma is expressed in terms of the multinomial coefficient d The following d! I := i1 !i2 !···im ! where I = (i1 , i2 , . . . , im ). Lemma 1.9.3. Let f ∈ K[V ]d be homogeneous of degree d, and consider the sequence of positive integers i1 , i2 , . . . , im with i1 + i2 + · · · + im = d. Then   d R(f(i1 ,i2 ,...,im ) ) = f . i1 i2 . . . im In particular, RP(f ) = d!f. Proof. Setting v = (w, w, . . . , w) we have λv = |λ|w where |λ| = λ1 + λ2 + · · · + λd . Therefore,  d f (λv) = f (|λ|w) = |λ|d f (w) = λI f (w) I |I|=d

Conversely. f (λv) =



λI fI (w, w, . . . w) =

|I|=d

Comparing the coefficients we see



λI RfI (w) .

|I|=d

d I f = RfI .

 

Remark 1.9.4. If d! is invertible in K, then f = RP(f /d!) lies in the image of G R. In particular, if f ∈ K[V ]G d and d is invertible in K, then f ∈ R(F[d V ] ). The following example illustrates polarization and restitution. Example 1.9.5. Let K be a field of any characteristic. Consider the usual three dimensional permutation representation V of Σ3 , the symmetric group on three letters. Let {x, y, z} be a permutation basis for V ∗ . It is well known that if K has characteristic zero, then K[V ]Σ3 is the polynomial ring K[s1 , s2 , s3 ] where s1 = x + y + z, s2 = xy + xz + yz and s3 = xyz. This result is also true when K has positive characteristic, even for characteristics 2 and 3. We will outline one proof of this result in §3.2 and give another proof in §5.1.1. Here we consider the ring of vector invariants K[2 V ]Σ3 . Weyl [112] proved that the polarizations of the elementary symmetric functions f = s1 , g = s2 , h = s3

14

1 First Steps

suffice to generate K[2 V ]Σ3 if 6 = |Σ3 | is invertible in K. In fact, he proved that if V is the usual permutation representation of Σn , then for any n and any m the polarizations of the elementary symmetric polynomials s1 , s2 , . . . , sn generate the ring K[m V ]Σn provided only that n! is invertible in K. Here we have f (λ1 v1 + λ2 v2 ) = f (λ1 x1 + λ2 x2 , λ1 y1 + λ2 y2 , λ1 z1 + λ2 z2 ) = λ1 (x1 + y1 + z1 ) + λ2 (x2 + y2 + z2 ). Thus Pol2 (f ) = {f10 , f01 } where f10 = x1 + y1 + z1 f01 = x2 + y2 + z2 . Similarly, g(λ1 v1 + λ2 v2 ) = g(λ1 x1 + λ2 x2 , λ1 y1 + λ2 y2 , λ1 z1 + λ2 z2 ) = λ21 (x1 y1 + x1 z1 + y1 z1 ) + λ1 λ2 (x1 y2 + x1 z2 + y1 x2 + y1 z2 + z1 x2 + z1 y2 ) + λ22 (x2 y2 + x2 z2 + y2 z2 ). Thus Pol2 (g) = {g20 , g11 , g02 } where g20 = x1 y1 + x1 z1 + y1 z1 , g11 = x1 y2 + x1 z2 + y1 x2 + y1 z2 + z1 x2 + z1 y2 , g02 = x2 y2 + x2 z2 + y2 z2 . Here g11 is the full polarization P(g). Finally h(λ1 x1 + λ2 x2 , λ1 y1 + λ2 y2 , λ1 z1 + λ2 z2 ) = λ31 (x1 y1 z1 ) + λ21 λ2 (x1 y1 z2 + x1 y2 z1 + x2 y1 z1 ) + λ1 λ22 (x1 y2 z2 + x2 y1 z2 + x2 y2 z1 ) + λ32 x2 y2 z2 . Hence Pol2 (h) = {h30 , h21 , h12 , h03 } where h30 = x1 y1 z1 , h21 = x1 y1 z2 + x1 y2 z1 + x2 y1 z1 , h12 = x1 y2 z2 + x2 y1 z2 + x2 y2 z1 , h03 = x2 y2 z2 . Weyl’s result tells us that if the characteristic of K is neither 2 nor 3, then K[2 V ]Σ3 is generated by the nine invariants f10 , f01 , g20 , g11 , g02 , h30 , h21 , h12 , h03 .

1.9 Polarization and Restitution

15

It turns out that these nine invariants also generate K[2 V ]Σ3 if K has characteristic 2. The identity 2 2 g02 − f10 f01 g11 + f10 h12 + g11 3(x1 y1 z22 + y1 z1 x22 + x1 z1 y22 ) = f10 2 − 2f10 h12 + f01 g11 − 4g20 g02 + 2f01 h21

shows how to express the invariant k := x1 y1 z22 + y1 z1 x22 + x1 z1 y22 in terms of the polarized elementary symmetric functions when 3 is invertible. However, over a field of characteristic 3, this identify expresses an algebraic relation among the polarized elementary symmetric functions. In fact, over a field of characteristic 3, it is not possible to express k as a polynomial in the nine polarized elementary symmetric functions. In fact, it turns out that the nine polarized elementary symmetric functions together with the invariant k form a minimal generating set for K[2 V ]Σ3 when K has characteristic 3. Polarization and restitution may be defined more generally, as follows. Given a multi-homogeneous function f ∈ K[W1 ⊕ W2 ⊕ · · · ⊕ Wt ](i1 ,i2 ,...,it ) and given positive integers m1 , m2 , . . . , mt , we define maps Δ : W1 ⊕ W2 ⊕ · · · ⊕ Wt → m1 W1 ⊕ m2 W2 ⊕ · · · ⊕ mt Wt and φ : m1 W1 ⊕ m2 W2 ⊕ · · · ⊕ mt Wt → W1 ⊕ W2 ⊕ · · · ⊕ Wt given by Δ(v1 , v2 , . . . , vt ) = (v1 , v1 , . . . , v1 , v2 , v2 , . . . , v2 , . . . , vt , vt , . . . , vt )          m1

m2

mt

and m1 m2 mt    v1j , v2j , . . . , vtj ). φ(v11 , v12 , . . . , v1m1 , . . . , vt1 , vt2 , . . . , vtmt ) = ( j=1

j=1

j=1

As above, these induce GL(W1 ) × GL(W2 ) × · · · × GL(Wt )-equivariant maps φ∗ : K[W1 ⊕ W2 ⊕ · · · ⊕ Wt ] → K[m1 W1 ⊕ m2 W2 ⊕ · · · ⊕ mt Wt ] and Δ∗ : K[m1 W1 ⊕ m2 W2 ⊕ · · · ⊕ mt Wt ] → K[W1 ⊕ W2 ⊕ · · · ⊕ Wt ]. Given f ∈ K[W1 ⊕ W2 ⊕ · · · ⊕ Wt ], the multi-homogeneous components of φ∗ (f ) are the partial polarizations of f . We denote the full set of these partial polarizations by Polm1 ,m2 ,...,mt (f ). As above, we distinguish as a special case, the full polarization of f . This is the unique multi-linear partial polarization and we again denote it by P(f ). The full polarization may also be described as follows. For each k = 1, 2, . . . , t, we let Pk : K[Wk ]dk → K[dk Wk ](1,1,...,1) denote the full polarization operator

16

1 First Steps

from the k th copy of V as defined earlier. Then we put P = Pt Pt−1 · · · P2 P1 which is given by P : K[W1 ⊕ W2 ⊕ · · · ⊕ Wt ](d1 ,d2 ,...,dt ) → K[d1 W1 ⊕ d2 W2 ⊕ · · · ⊕ dt Wt ](1,1,...,1) . It is easy to see that these more general partial polarization operators are GL(W1 ) × GL(W2 ) × · · · × GL(Wt ) − equivariant and that P(f ) is symmetric (invariant) under the action of Σd1 × Σd2 × · · · × Σdt . We define a generalized restitution operator R = R(r1 ,r2 ,...,rt ) = Rt ◦ Rt−1 ◦ · · · ◦ Rt ◦ R1 similarly: R : K[r1 W1 ⊕ r2 W2 ⊕ · · · ⊕ rt Wt ] → K[W1 ⊕ W2 ⊕ · · · ⊕ Wt ]; so that R(F )(v1 , . . . , vt ) = F (v1 , . . . , v1 , v2 , . . . , v2 , . . . , vt , . . . , vt )          r1

r2

rt

for F ∈ K[r1 W1 ⊕ r2 W2 ⊕ · · · ⊕ rt Wt ]. Then R(P(f )) = d1 !d2 ! · · · dt !f if f ∈ K[W1 ⊕W2 ⊕· · ·⊕Wt ](d1 ,d2 ,...,dt ) . Note that unlike polarization, restitution is an algebra homomorphism.

1.10 The Role of the Cyclic Group Cp in Characteristic p In many respects, the characteristic p invariant theory of the cyclic group Cp of order p plays a central role in modular invariant theory. In this book, we will spend considerable effort developing our understanding of Cp -invariants in characteristic p. To partially explain the importance of Cp , we begin with the following two very useful lemmas. Lemma 1.10.1. Suppose H is a normal subgroup of G with quotient group G/H. Let V be a representation of G. Then G/H acts naturally on V H and V G = (V H )G/H .   We will use Lemma 1.10.1 in the proof of the next lemma. However, its main use will be when we apply it to a normal subgroup H of a group G acting on a coordinate ring K[V ]. Then we have K[V ]G = (K[V ]H )G/H . This is the topic of Chapter 14. Lemma 1.10.2. Let G be any p-group for p a prime and let H be any maximal proper subgroup. Then H is normal in G necessarily of index p. Hence if G is generated by H and σ, we have G/H = Cp generated by σ ¯ , the image of σ in G/H.   The preceding lemma shows that for any p-group G, we may construct a composition series, that is, construct a tower of groups Gi , each normal in the next such that Gi+1 /Gi ∼ = Cp with G0 = {e} and Gm = G. We record this result as the following lemma.

1.11 Cp Represented on a 2 Dimensional Vector Space in Characteristic p

17

Lemma 1.10.3. Suppose G is a p-group. Then G is solvable with all composition factors isomorphic to Cp . That is, {e} = G0 G1 G2 . . . Gm = G with Gi /Gi−1 ∼ = Cp for all i = 1, 2, . . . , m.

 

A consequence of this lemma is that we may compute the invariants of a p-group G by repeatedly computing invariants under an action of the cyclic group Cp . Given {e} = G0 G1 G2 . . . Gm = G, we proceed as follows. First, we compute R1 := K[V ]G1 where G1 ∼ = Cp . Then we compute R2 = C K[V ]G2 = (K[V ]G1 )G2 /G1 ∼ = R1 p . Continuing in this manner we compute C Rj+1 = K[V ]Gj+1 = (K[V ]Gj )Gj+1 /Gj ∼ = Rj p for j = 0, 1, . . . , m. This yields Rm+1 = K[V ]G . Thus, in theory at least, any composition series of G provides an inductive method of computing the G-invariants. Of course, this so-called “ladder method” is applicable to any solvable group. In practice, this method runs into difficulties particularly for representations of G over a field K of characteristic p, see §14. Heuristically, the problems occur because Cp is acting on K[V ]H which is most often not polynomial and, in particular, is not of the form K[W ]. For modular representations, this presents special difficulties. Chapter 14 discusses this technique in detail and shows how we may use group cohomology to handle the extra difficulties that arise in the modular case.

1.11 Cp Represented on a 2 Dimensional Vector Space in Characteristic p As a simple example of a modular group action, consider the vector space V2 of dimension 2 over a field F of characteristic p > 0 with basis {e1 , e2 }. We start with a lengthy but elementary proof which illustrates part of the attraction of modular invariant theory. Namely, that it is possible to prove some theorems using only basic techniques. Let Cp denote the cyclic group of order p generated by σ. Consider the matrix   10 τ= 11 inside GL(2, F) where F is a field of characteristic p. It is easy to show, using induction, that   10 i τ = . i1 Therefore, we obtain a representation ρ : Cp → GL(V2 ) given by the rule ρ(σ i ) = τ i . We have σ(e1 ) = τ (e1 ) = e1 + e2 and σ(e2 ) = τ (e2 ) = e2 . Let {x, y} be the basis for V2∗ dual to {e1 , e2 }. Then σ(x) = x and σ(y) = −x + y.

18

1 First Steps

We see immediately that the polynomial x is an invariant. Moreover, since (y + x)p = y p + xp , the polynomial N = y p − xp−1 y is another example of an invariant. We will see below, in Theorem 1.11.2, that for this representation, these two invariants are the two most important invariants. Lemma 1.11.1. NCp (y) = y p − xp−1 y.

 

Our goal is to show that the ring of Cp -invariants is the algebra generated by the two invariants N and x: Theorem 1.11.2. F[V2 ]Cp = F[x, N ]. Before proving Theorem 1.11.2, we need some preliminary results. Lemma 1.11.3. Let f ∈ F[V2 ]. Then degy (σ(f )) = degy (f ). Proof. Let m denote degy (f ) and write f = am y m + am−1 y m−1 + · · · + a0 where ai ∈ F[x] and am = 0. Then σ(f ) = σ(am )(σ(y)m ) + σ(am−1 )(σ(y)m−1 ) + · · · + σ(a0 ) = am (y − x)m + am−1 (y − x)m−1 + · · · + a0 = am y m + terms of lower order in y  

Thus degy (σ(f )) = m.

Since N is monic when considered as a polynomial in the variable y with coefficients from F[x], we may divide any polynomial f ∈ R by N to get f = qN + r where q, r ∈ F[x, y] are unique with degy (r) < p = degy (N ). Lemma 1.11.4. If f ∈ F[V2 ]G and f = qN + r with degy r < p, then q, r ∈ F[V2 ]G . Proof. First we note that it is enough to show that q and r are σ-invariant since σ generates Cp . We have f = σ(f ) = (σ · q)(σ(N )) + (σ · r) = (σ · q)N + (σ · r). Since degy (σ · r) = degy (r) < p, by the uniqueness of remainders and quotients we must have σ · r = r and σ · q = q.   Now we need a result concerning the partial differential operator Lemma 1.11.5. If f ∈ F[x, y]G , then

∂ ∂ y (f )

∂ ∂y.

∈ F[x, y]G .

Proof. We note that it is sufficient to show that if f ∈ F[V ], then σ( ∂∂y (f )) = ∂ ∂ ∂ y (σf ). Further, we note that both σ and ∂ y are F-linear maps. Therefore, to show that they commute we need only show that they commute on monomials: ∂ a b (x y )) = σ(bxa y b−1 ) = bxa (y − x)b−1 and ∂y ∂ ∂ a (σ · xa y b ) = (x (y − x)b ) = bxa (y − x)b−1 ∂y ∂y

σ(

 

1.11 Cp Represented on a 2 Dimensional Vector Space in Characteristic p

19

Remark 1.11.6. The lemma above also follows from the Leibnitz’ rule: ∂ ∂ ∂ (f1 f2 ) = (f1 )f2 + f1 (f2 ) ∂y ∂y ∂y We now give the proof of Theorem 1.11.2. Proof. Clearly, F[V2 ]Cp ⊇ F[N, x]. Thus it suffices to prove that each invariant, f , is contained in F[x, N ]. We prove this by induction on degy (f ). If degy (f ) = 0, then f ∈ F[x] ⊂ F[x, N ]. Next, suppose degy (f ) = d and that every invariant, whose degree in y is less than d, lies in F[x, N ]. Write f = q · N + r, where m := degy (r) < p. We will now show m = 0. Assume, by way of contradiction, that m ≥ 1 and consider the invariant h defined by h :=

∂ m−1 (r) . ∂y m−1

Then h = ay + b, where a is a non-zero scalar and b ∈ F[x]. But h = σ(ay + b) = a(y − x) + b = ay + b − ax and this contradiction shows that we must have m = 0. Therefore, f = q · N + r where q ∈ F[V2 ]Cp , degy (q) = d − p and r ∈ F[x]. By the induction hypothesis, q ∈ F[x, N ] and thus f ∈ F[x, N ].   In later chapters, we will discuss some elements of commutative algebra and we will be able to give a simpler proof of this result. An outline of this simpler proof is as follows. Since {x, N } is a homogeneous system of parameters for F[V2 ]Cp (see §2.6 for details) and the product of their degrees equals the order of the group, then Theorem 1.11.2 follows from Theorem 3.1.6. We pause here for a discussion of history and philosophy. The invariant theory of polynomials began in characteristic 0. For example, at the time of Newton and Vandermonde, there was intense interest in generalizing the famous quadratic formula. The problem was to find the roots of high degree polynomials in one variable with integer coefficients by means of radicals. One method of study was to suppose the solution and study which polynomials arise: perhaps, it was thought, all of them. Let x1 , . . . , xn denote the roots of a polynomial f (t) = an tn + · · · + a1 t + a0 for integers ai . This means, of course, that f (t) = an

n 

(t − xi )

i=0

and a0 , a1 , . . . , an−1 are the elementary symmetric polynomials in x1 , x2 , . . . , xn . We note that the right hand side is invariant under any permutation of the variables. Therefore, in order to understand solutions of such equations by means of radicals it is possibly useful to understand the invariants of permutations of n-variables. Such considerations lead to the invariant theory of

20

1 First Steps

other groups, and many books on invariant theory begin with this particular situation. For another approach, we note that Theorem 1.11.2 tells us that the ring of invariants F[V2 ]Cp is a polynomial algebra. Characterizing the modular groups with this property, where the rings of invariants are again polynomial algebras, is still an open question, perhaps the most important open problem in this area of research. By way of contrast, the characterization of such groups in the non-modular case (Theorem 1.5.2) is one of the best-known and beautiful results in classical invariant theory. We are attracted to such results because an important property of the original algebra is preserved under the action of the group. It is not difficult to show that F(V2 )Cp and F(x, N) are equal using Galois Theory. It is often not too difficult to discover sets of invariants with the property that the quotient field they generate is the quotient field of the ring of invariants. Heuristically, as one is led to believe by Noether’s question, fewer generators are needed at the quotient field level to generate the field of invariants. In any event, Theorem 1.11.2 is attractive also in this sense — that the generators needed for the field of invariants suffice to generate the ring of invariants as well. The example, while simple and straightforward, offers a glimpse into the world of invariant theory. We seek to discover which invariants generate the ring of invariants and we are interested in the algebraic structures that are exhibited by the invariant ring.

1.12 A Further Example: Cp Represented on 2 V2 in Characteristic p Here we compute the ring of invariants of the group Cp on 2 V2 := V2 ⊕ V2 . Here we are considering the action of Cp determined by ⎛ ⎞ 1000 ⎜1 1 0 0⎟ ⎟ σ=⎜ ⎝0 0 1 0⎠ . 0011 We introduce

and

⎛

⎞ 1 000 ⎜−1 1 0 0⎟ ⎟ σ1 = ⎜ ⎝ 0 0 1 0⎠ 0 001 ⎛

1 ⎜0 σ2 = ⎜ ⎝0 0

0 1 0 0

0 0 1 1

⎞ 0 0⎟ ⎟. 0⎠ 1

1.12 A Further Example: Cp Represented on 2 V2 in Characteristic p

21

Thus σ = σ2 σ1−1 . Let {x1 , y1 , x2 , y2 } be the basis for the vector space (2 V2 )∗ dual to the standard basis of 2 V2 . Thus σi (xj ) = xj for 1 ≤ i, j ≤ 2, σi (yj ) = yj for 1 ≤ j = i ≤ 2, σ1 (y1 ) = y1 + x1 and σ2 (y2 ) = y2 − x2 . Define G to be the group generated by σ1 and σ2 , so that G = Cp × Cp and H to be the group generated by σ = σ2 σ1−1 so that H = Cp . We want to compute F[2 V2 ]H . Given the example just computed, it is easy to see that F[2 V2 ]G = F[V2 ]Cp ⊗ F[V2 ]Cp = F[x1 , N (y1 ), x2 , N (y2 )]. Now we consider the diagram

→

→

→

F(2 V2 )G → F(2 V2 )H → F(2 V2 )

F[2 V2 ]G → F[2 V2 ]H → F[2 V2 ]. By Galois Theory, we have that the field F(2 V2 )H is an extension of the field F(2 V2 )G of degree p; that is, F(2 V2 )H as vector space over F(2 V2 )G has dimension p. In order to exploit this property, we need to find an element of F(2 V2 )H that lies outside of F(2 V2 )G . It is easy to see that the element u = x1 y2 − x2 y1 is an invariant of least degree in F[2 V2 ]H outside F[2 V2 ]G , and therefore F(2 V2 )H has basis   1, u, u2 , . . . , up−1 . We see that xp−1 u. up = xp1 N (y2 ) − xp2 N (y1 ) + xp−1 1 2 Theorem 1.12.1. F[2 V2 ]Cp = F[x1 , x2 , N(y1 ), N(y2 ), u] . In fact, i F[2 V2 ]Cp = ⊕p−1 i=0 F[x1 , x2 , N(y1 ), N(y2 )]u .   Proof. Our challenge is to show that 1, u, u2 , . . . , up−1 is a basis for F[2 V2 ]H as a module over F[2 V2 ]G . Since G is Abelian, we have that H is normal in G and hence

F[2 V2 ]G = (F[2 V2 ]H )G/H . We note that the image of either σ1 or σ2 generates G/H = Cp . We define

22

1 First Steps

Δ1 = σ1 − Id Δ2 = σ2 − Id and Δ = σ − Id. Consider f ∈ F[2 V2 ]H and note that σ(f ) = f implies σ1 (f ) = σ2 (f ) and thus Δ1 (f ) = Δ2 (f ). In particular, f ∈ F[2 V2 ]G if and only if Δ1 (f ) = 0. Also σ(Δ1 (f )) = σ2 σ1−1 (σ1 (f ) − f ) = σ2 (f ) − σ(f ) = σ1 (f ) − f = Δ1 (f ). Thus Δ1 : F[2 V2 ]H → F[2 V2 ]H . Lemma 1.12.2. If f ∈ F[2 V2 ]H , then Δ1 (f ) = x1 x2 f  for some f  ∈ F[2 V2 ]H . d Proof. For any f ∈ F[2 V2 ] we write f = =0 f y1 with f ∈ F[x1 , x2 , y2 ] for 0 ≤ ≤ d in order to see that Δ1 (f ) = x1 f  . Similarly, Δ2 (f ) = x2 f  . If f ∈ F[2 V2 ]H , then σ1 (f ) = σ2 (f ), and so x1 f  = x2 f  . But x1 and x2 are co-prime in F[2 V2 ] and so Δ1 (f ) = x1 x2 f  for some f  ∈ F[2 V2 ]. Since both   Δ1 (f ) and x1 x2 are H-invariant, we see that f  ∈ F[2 V2 ]H . We now finish the proof of Theorem 1.12.1. Since Δp1 = (σ1 − Id)p = σ1p − Id = 0 we see that Δp1 (f ) = 0 for all f ∈ F[2 V2 ]. Thus given 0 = f ∈ F[2 V2 ]H there must exist an , 0 ≤ < p with the property that 0 = Δ1 (f ) ∈ F[2 V2 ]H  m G and Δ+1 1 (f ) = 0. We claim that then f = m=0 fm u for fm ∈ F[2 V2 ] . We proceed by induction on . If = 0 then Δ1 (f ) = 0 which implies f ∈ F[2 V2 ]G as we observed above. For the general case, we write Δ1 (f ) = x1 x2 f  with −1  m  u with all fm ∈ F[2 V2 ]G by f  ∈ F[2 V2 ]H and observe that f  = m=0 fm induction. Now consider Δ1 (f

  −1 1   m+1 + uf / ) = f u f+

m=0 m   −1 1   m+1 (f ) + f Δ (u ) Δ = Δ−1 1 1 1

m=0 m   −2 1  1     m+1 x f + Δ (u ) + f Δ (u ) = Δ−1 x f 1 2 1 1 1

−1

m=0 m  −1   −1 

i −1  m1  x1 x2 f m u = Δ1 u (−x1 x2 )−i f−1 i

m=0 i=0  −2  1 + f  Δ1 (um+1 )

m=0 m 

Δ1

1.13 The Vector Invariants of V2

23

−2   1   i = + f−1 u (−x1 x2 )−i i

m=0 i=0  −2  1 + f  Δ1 (um+1 )

m=0 m   −2 m h u = Δ−1 m 1

Δ−1 1

 −2

 m x1 x2 fm u

m=0

where hm ∈ F[2 V2 ]G for m = 1, 2, . . . , − 2 and this final expression is equal to zero since, as is easily verified, Δt1 (us ) = 0 whenever t > s. G m by the induction hypothesis. Thus Therefore, f + uf  / ∈ ⊕−1 m=0 F[2 V2 ] u  G m f ∈ ⊕m=0 F[2 V2 ] u which proves Theorem 1.12.1.  

1.13 The Vector Invariants of V2 Given a representation V of a group G and an integer m ≥ 2, a ring of invariants K[m V ]G is called a ring of vector invariants. A theorem providing an explicit description of K[m V ]G for all m ≥ 1 is called a first fundamental (or main) theorem for V . The following first fundamental theorem for V2 was conjectured by David Richman and proved by Campbell and Hughes, see [19]. Their proof is technical and uses a deep result about the rank of zero-one matrices in characteristic p. We will give a shorter proof which uses ideas from this book and which has the advantage that it yields more than just a generating set; it yields a SAGBI basis as we shall see in §7.4. Theorem 1.13.1. Let G = Cp = σ act on V = m V2 . Let {yi , xi } denote a basis for the ith copy of V2∗ in V ∗ where σ(yi ) = yi + xi and σ(xi ) = xi . Thus {x1 , y1 , x2 , y2 , . . . , xm , xm } is an upper triangular basis for V ∗ . Then the ring of invariants F[m V2 ]Cp is generated by the following invariants: 1. xi for i = 1, 2, . . . , m. 2. NCp (yi ) = yip − xip−1 yi for i = 1, 2, . . . , m. 3. uij = xj yi − xi yj for 1 ≤ i < j ≤ m. am ) where 0 ≤ ai < p for i = 1, 2, . . . , m. 4. TrCp (y1a1 y2a2 . . . ym Remark 1.13.2. Shank and Wehlau [99] showed that if a1 +a2 +· · ·+am ≤ 2(p− am 1), then TrCp (y1a1 y2a2 . . . ym ) lies in the subalgebra generated by x1 , x2 , . . . , xm and uij with 1 ≤ i < j ≤ m. Additionally, they also showed that if we exclude invariants of this form, the remaining invariants minimally generate F[m V2 ]Cp . The following example illustrates Theorem 1.13.1. Example 1.13.3. If we take m = 3 and F a field of characteristic p = 3, then Theorem 1.13.1 tells us that F[3 V2 ]C3 is generated by x1 , x2 , x3 , N(y1 ), N(y2 ), N(y3 ), u12 , u13 , u23 and some transfers.

24

1 First Steps

It is straightforward to compute TrC3 (yi ) = 0 Tr

C3

C3

Tr Tr

C3

Tr

for i = 1, 2, 3;

(yi yj ) = −xi xj

(yi2 yj )

for 1 ≤ i, j ≤ 3; for 1 ≤ i = j ≤ 3;

= xi uji

(y1 y2 y3 ) = x1 u23 − x3 u12 ;

C3

(yi2 yj2 ) = −uij − x2i x2j

Tr(yi y1 y2 y3 ) = −uij uik −

x2i xj xk

for 1 ≤ i < j ≤ 3; where {i, j, k} = {1, 2, 3}.

Thus we see, in agreement with Remark 1.13.2, that F[3 V2 ]C3 is minimally generated by x1 , x2 , x3 , N(y1 ), N(y2 ), N(y3 ), u12 , u13 , u23 , TrC3 (y12 y22 y3 ), TrC3 (y12 y2 y32 ), TrC3 (y1 y22 y32 ) and TrC3 (y12 y22 y32 ). Remark 1.13.4. The proof of Theorem 1.13.1 (see §7.4) shows in particular p−1 that the invariant TrCp (y1p−1 y2p−1 . . . ym ) cannot be expressed using only invariants of lower degree and thus the Noether β(m V2 , Cp ) ≥ m(p − 1). (Of course, the theorem also shows that we have equality here.) Similarly, in Corollary 7.7.3, we show that (over a field of characteristic p) β(m Vr , Cp ) > m(p−1) if r ≥ 3. Thus, unlike the non-modular situation, in the modular setting, there can be no general bound on β(V, G) independent of V . In fact, fix any field K and any linear algebraic group G and consider β(G) := sup{β(V, G) | G ≤ GLK (V )}. Bryant and Kemper [15] showed that β(G) is finite if and only if G is a finite group whose order is invertible in K.

2 Elements of Algebraic Geometry and Commutative Algebra

In this chapter we summarize the basic elements of algebraic geometry and commutative algebra that are useful in the study of (modular) invariant theory. Normally, these techniques are most useful in questions about the structure of rings of invariants, and, as well, they seem to be most useful in proving theorems that hold true for all groups, modular or not. It is worth noting that a large fraction (as much as one third — see the paper of Fisher [37, Page 146]) of the papers in mathematics in the latter stages of the 19th century were studies of invariant theory. It is worth noting as well that commutative algebra was invented, discovered if you prefer, by Hilbert, in order to clarify and understand invariant theory more fully, see Fisher [37]. There are several excellent references, including Atiyah and Macdonald [5], Dummit and Foote [32], Eisenbud [33], Lang [74], Matsumura [79], and a forthcoming book by Kemper, [67].

2.1 The Zariski Topology Algebraic geometry uses algebra to study geometric objects and vice versa. Given a vector space V defined over a field K we study the K-valued polynomial functions on V . Let {x1 , x2 , . . . , xn } be a basis for V ∗ = homK (V, K), the vector space of linear functionals on V . We consider the polynomial ring S = K[x1 , x2 , . . . , xn ]. Here each xi is naturally a (linear) function on V and we consider a polynomial in S to be a function on V by using the usual pointwise definitions for products and sums of functions. Thus each element of f ∈ S is a K-valued function, f : V → K with domain V . We would like to use the geometry of V and the action of G on V to study S and its structure as a G-module. However, there is a problem for us since we will most often be working over a finite field F. If F is finite then so is V and thus there are only finitely many functions from V to F. However, the polynomial ring S is always infinite. For example, if F = Fp , the field of order p, then the two polynomials x1 and xp1 represent the same function from V to H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 2, © Springer-Verlag Berlin Heidelberg 2011

25

26

2 Elements of Algebraic Geometry and Commutative Algebra

Fp . There are a few ways around this difficulty. We will proceed by extending our field K to an algebraically closed field K. We will work with both K and K and then interpret what our results say about invariants over K. We write V := V ⊗K K. We define the coordinate ring of V by K[V ] := K[x1 , x2 , . . . , xn ] and the coordinate ring of V by K[V ] := K[x1 , x2 , . . . , xn ]. In this section, the connection of this coordinate ring K[V ], to the geometry of V , is explored. Given a set T of polynomials in S we define V(T ) = V V (T ) to be the subset of V on which every element of T vanishes, that is, V V (T ) = V(T ) = {v ∈ V | f (v) = 0 for all f ∈ T }. We say that V(T ) is the variety that is “cut out” by the set T . A subset X of V which can be realized as X = V(T ) for some subset T of S is called an algebraic subset of V . Given such an algebraic subset X of V we may view functions on V as functions on X via restriction. By definition, the coordinate ring of the subvariety X is the set of such restricted functions. Thus restriction gives a surjective algebra map, res : K[V ] → K[X]. Hence K[X] ∼ = K[V ]/I(X) where the ideal I(X) is the kernel of res. Thus I(X) = IS (X) is the subset of polynomials in S vanishing on X, that is, IS (X) = I(X) = {f ∈ S | f (v) = 0 for all v ∈ X}. The following celebrated theorem of Hilbert holds, see [5, pg 85]. . . . , xn ], where Theorem 2.1.1. (Hilbert’s Nullstellensatz) Let S = K[x1 , x2 ,√ K is algebraically closed. If I is an ideal of S then I(V(I)) = I. The following two results will be useful later on. Lemma 2.1.2. Let v1 , v2 , . . . , vm be distinct elements of the K-vector space V . Then there exists a (non-homogeneous) h ∈ K[V ] such that  1 if i = 1, h(vi ) = 0 if i = 1 Corollary 2.1.3. Let v1 , v2 , . . . , vm be distinct elements of the K-vector space V . Let c1 , c2 , . . . , cm ∈ K be given. Then there exists f ∈ K[V ] such that f (vi ) = ci for i = 1, 2, . . . , m. The concept of the coordinate ring of V can be extended to any algebraic subset X ⊆ V by considering the functions on V restricted to X. Of course, any function of I(X) when restricted to X gives the zero function. Thus two functions f1 , f2 ∈ S = K[V ] have the same restriction to X if and only if f1 − f2 ∈ I(X). Thus studying the restrictions of elements of S to X is equivalent to studying the ring S/I(X). This ring, S/I(X), is called the coordinate ring of the subset X and is denoted K[X].

2.3 Noetherian Rings

27

We note that K[X] will fail to be a domain when I(X) is not prime, i.e., when X is not irreducible. However, K[X] does not have nilpotent elements because I(X) is a radical ideal. In fact, as we shall see in the next section, every Noetherian K algebra without nilpotent elements can be realized as the coordinate ring of some variety X.

2.2 The Topological Space Spec(S) Given a coordinate ring S = K[V ] as above, we consider two sets; MaxSpec(S) denotes the set of all maximal ideals of S, and Spec(S) denotes the set of all prime ideals of S. Minimal closed subsets of V correspond to the maximal ideals of S so that there is a one-to-one correspondence between the points of V and the elements of MaxSpec(S). This correspondence identifies V with MaxSpec(S). Under this identification, the Zariski topology on V makes MaxSpec(S) into a topological space. This topology passes to Spec(S) as well: M ⊂ Spec(S) is closed if and only if there exists an ideal J of S such that M consists of all the prime ideals of S which contain J, i.e, if and only if M = {I ∈ Spec(S) | I ⊇ J}. Hence a point of the topological space Spec(S), i.e., a prime ideal of S, is closed in this topology if and only if it is a maximal ideal of S. This topology on Spec(S) is also referred to as the Zariski topology. We now return to the question of non-algebraically closed fields. Often, our field K is not algebraically closed, for example, it is often a finite field. Since our main object of study is K[V ] = K[x1 , x2 , . . . , xn ], we prefer not to merely extend the base field and study K[V ]. Instead, we compromise and consider the elements of K[x1 , x2 , . . . , xn ] as functions from V to K, i.e., we view K[V ] as a sub-ring of K[V ]. With this bigger domain, each of the (infinitely many) elements of K[V ] corresponds to a different function on V . This allows us to treat the elements of K[V ] solely as functions and so to relate the algebraic properties of K[V ] to the geometric properties of V .

2.3 Noetherian Rings We will always study rings that are commutative with a 1. There are many good references including Atiyah and Macdonald [5], Bruns and Herzog [14], Dummit and Foote [32], Eisenbud [34], Land [75], Matsumura [80], Zariski and Samuel [116] and [117]. We assume that R is non-negatively graded, that is R = ⊕∞ d=0 Rd where each Rd is a K vector space and Ri · Rj ⊆ Ri+j for all i, j ∈ N. We further suppose that R is connected, i.e., that R0 = K. Finally, we will also assume that R is finitely generated. This means that there is a finite collection of elements Z = {f1 , f2 , . . . , ft } ⊂ R such that every element of R may be written as a K linear combination of products whose factors are elements

28

2 Elements of Algebraic Geometry and Commutative Algebra

of Z: R = K[f1 , f2 , . . . , ft ]. If R is a graded connected finitely generated Kalgebra and all of the generators f1 , f2 , . . . , ft may be taken in R1 then we say that R is a standard graded K-algebra. We note that a finitely generated K-algebra is always Noetherian, i.e., every ideal I of R is finitely generated: I = (g1 , g2 , . . . , gs ) = Rg1 + Rg2 + · · · + Rgs . An element h ∈ R is homogeneous (of degree d) if there exists d such that h ∈ Rd . An ideal I of R is graded (or homogeneous) if it generated by homogeneous elements. Equivalently, I is graded if whenever h ∈ I and we write h = h0 + h1 + · · · + hd with hi ∈ Ri , we have all h1 , h2 , . . . , hd ∈ I. The ideal R+ := ⊕∞ d=1 Rd is the unique maximal homogeneous ideal in R. The prototypical example of a finitely generated K-algebra is the polynomial ring K[x1 , x2 , . . . , xn ], where the xi are (algebraically independent) indeterminates. If I ⊂ R is a graded ideal then the quotient ring R/I is also a graded ring via (R/I)d = Rd /Id . Furthermore, if R is a standard graded algebra, then R/I is also standard graded. In fact, it can be shown that every standard graded K-algebra arises in this manner. Suppose R is a finitely generated K-algebra which is a domain. Let f1 , f2 , . . . , fm denote the set of generators for R. Consider the polynomial ring S = K[x1 , x2 , . . . , xm ]. We may view S as the coordinate ring of the K-vector space V of dimension m, i.e., S = K[V ] where we identify the indeterminates x1 , x2 , . . . , xm with a basis of V ∗ . Then we have a surjection π:S→R given by π(xi ) = fi . Let I denote the kernel of π. Since we have assumed that R is a domain, I is a prime ideal of S. Thus we may view R as the coordinate ring of an (irreducible) algebraic subset X of V . Here X = V V (I). The map of varieties which is dual to the ring π is an embedding of X into V ∼ = Km . For this reason, if R is minimally generated by m elements we call m the embedding dimension of R, denoted edim(R). Let R be a Noetherian ring and suppose ℘ is a prime ideal in R. We define the height of ℘ to be i if ℘0  ℘1  · · ·  ℘i = ℘ is a chain of prime ideals of maximal length of R contained in ℘. More generally, if I is any ideal of R, then the height of I is defined by height(I) = min {height(℘) | ℘ ⊇ I, ℘ is prime}. Finally, the Krull dimension of R is the supremum of the heights of all prime ideals in R. Example 2.3.1. A polynomial algebra K[x1 , x2 , . . . , xn ] has Krull dimension n. Of course, the embedding dimension of a finitely generated K algebra R is always greater than or equal to its Krull dimension with equality if and only R is a polynomial ring. A Noetherian ring R is called a hypersurface ring or just a hypersurface if edim(R) ≤ Krull dim(R) + 1.

2.5 Integral Extensions

29

2.4 Localization and Fields of Fractions Let R be a commutative ring and let T ⊂ R be a subset which is closed under multiplication and which contains 1. We introduce an equivalence relation ∼ on R × T by putting (r, t) ∼ (r , t ) ⇐⇒ s(rt − r t) = 0 for some s ∈ T . We write the fraction a/b to denote the equivalence class of (a, b) and T −1 R to denote (R × T )/ ∼ which is called the localization of R at T . The operations a/b + c/d := (ad + bc)/bd and (a/b)(c/d) = ac/bd endow T −1 R with a ring structure. An important example of this construction is T = {f m | m ∈ N} where f is some element of R which is not a zero divisor. In this situation we write Rf to denote T −1 R. Another important example arises from a prime ideal ℘ in R. Since ℘ is prime, its complement T = R \ ℘ is a multiplicatively closed set containing 1. We write R℘ for T −1 R. In our setting, the natural map R → R℘ carrying a to a1 is injective. Furthermore, there is a bijection (induced by the map R → R℘ ) between the set of prime ideals of R℘ and the set of prime ideals of R which are contained in ℘. One of the most important features of localization at ℘ is that R℘ is a local ring, i.e., R℘ has a unique maximal ideal. If R is a domain, then (0) is a prime ideal. Then R(0) is just the field of fractions or quotient field of R, which we denote by Quot(R). An important property of quotient fields we will exploit is the following. Suppose that R ⊆ S are domains which are finitely generated algebras over the same coefficient field, and that R and S share the same Krull dimension. Then Quot(S) is a finite dimensional Quot(R) vector space.

2.5 Integral Extensions Suppose that R is a subring of S. An element y of S is integral over R if there is a monic polynomial f (X) = X r + fr−1 X r−1 + · · · + f0 with f1 , f2 , . . . , fr ∈ R such that f (y) = 0. We say that S is an integral extension of R or simply that S is integral over R if every element y of S is integral over R. The integral closure of R in S is the set of elements of S which are integral over R. The normalization of a domain R is its integral closure in its field of fractions Quot(R). A domain R is integrally closed or normal if it equals its normalization. Proposition 2.5.1. Let R be a subring of S and suppose y ∈ S. Then y is integral over R if and only if the ring R[y] is a finitely generated R-module. Furthermore, if S is a finitely generated R-algebra, then S is integral over R if and only if S is a finitely generated R-module.

30

2 Elements of Algebraic Geometry and Commutative Algebra

We will use the following well-known theorem many times. Theorem 2.5.2. Suppose that S is an integral extension of R. 1. Lying Over: Let p be a prime ideal of R. Then there exists a prime ideal q of S with q ∩ R = p. Moreover, p is a maximal ideal if and only if q is. We say that q lies over p. 2. Going-Up: Let p1  p2  · · ·  pt be an ascending chain of prime ideals in R. Suppose t > s and q1  q2  · · ·  qs is an ascending chain of prime ideals in S with qi ∩ R = pi for i = 1, 2, . . . , s. Then the ascending chain of primes in S can be extended, i.e., there exist prime ideals qs+1 , . . . , qt of S such that q1  q2  · · ·  qt and qi ∩ R = pi for all i = 1, 2, . . . , t. 3. Going-Down: Further suppose that R is integrally closed. Let p1  p2  · · ·  pt be a descending chain of prime ideals in R. Suppose t > s and q1  q2  · · ·  qs is a descending chain of prime ideals in S with qi ∩ R = pi for i = 1, 2, . . . , s. Then the descending chain of primes in S can be extended, i.e., there exist prime ideals qs+1 , . . . , qt of S such that q1  q2  · · ·  qt and qi ∩ R = pi for all i = 1, 2, . . . , t.

2.6 Homogeneous Systems of Parameters Let R be a finitely generated K algebra. A homogeneous system of parameters of positive degree for R is a set of homogeneous elements {f1 , f2 , . . . , fn } of R where n is the Krull dimension of R and with the property that R is finitely generated as a module over the ring A = K[f1 , f2 , . . . , fn ]. Equivalently, R is integral over K[f1 , . . . , fn ]. The Noether Normalization Lemma, a fundamental theorem of modern algebra, asserts that homogeneous systems of parameters always exist. This result is used repeatedly in the study of the polynomial invariants of finite groups. Noether’s Normalization Lemma was originally proved by Hilbert but has acquired its name since Emmy Noether stated and proved it as a lemma in her 1926 paper [88] in order to prove her main result that rings of invariants of finite groups are finitely generated algebras (Theorem 3.1.2). Theorem 2.6.1 (Noether Normalization Lemma). Let A be a finitely generated graded connected K-algebra. Then A has a homogeneous system of parameters. Remark 2.6.2. If A is a finitely generated graded K-algebra of Krull dimension n, and {h1 , h2 , . . . , hn } is a (any) homogeneous system of parameters, then we will refer to B = K[h1 , h2 , . . . , hn ] as a Noether Normalization of A. A sequence h1 , h2 , . . . , ht in R is a partial homogeneous system of parameters if it can be extended to a homogenous system of parameters h1 , h2 . . . , ht , ht+1 , . . . , hn . The following lemma is a very useful characterization of homogeneous systems of parameters.

2.7 Regular Sequences

31

Lemma 2.6.3. Let A ⊂ K[V ] be a Noetherian graded subring of Krull dimension n = dim V . Suppose {h1 , h2 , . . . , hn } is a set of homogeneous elements of A. Then h1 , h2 , . . . , hn is a homogeneous system of parameters for A if and only if V V (h1 , h2 , . . . , hn ) = {0}. Proof. An ideal (f1 , f2 , . . . , ft ) generated by t (homogeneous) elements has height less than or equal to t with equality if and only if (f1 , f2 , . . . , ft ) is a partial (homogeneous) system of parameters. Thus h1 , h2 , . . . , hn is a homogeneous system of parameters if and only if height(h1 , h2 , . . . , hn ) = n if and only if dim V(h1 , h2 , . . . , hn ) = 0 if and only if V V (h1 , h2 , . . . , hn ) is a finite set of points. Now since each hi is homogeneous, V V (h1 , h2 , . . . , hn ) is stable under scaling, i.e., if v ∈ V V (h1 , h2 , . . . , hn ) then λv ∈ V V (h1 , h2 , . . . , hn ) for all λ ∈ K. In particular, if V V (h1 , h2 , . . . , hn ) contains any point other than the origin then V V (h1 , h2 , . . . , hn ) contains a line and is thus infinite. This shows that V V (h1 , h2 , . . . , hn ) is finite if and only if V V (h1 , h2 , . . . , hn ) = {0}.  

2.7 Regular Sequences A sequence {r1 , . . . , rs } in R is called a regular sequence if (r1 , . . . , rs )R = R and ri is not a zero-divisor in R/(r1 , . . . , ri−1 )R for 1 ≤ i ≤ s. (For i = 1 this means r1 is not a zero-divisor in R.) We call s the length of the sequence. If R is a Noetherian ring and r1 , r2 , . . . is any regular sequence in R then the sequence of ideals R(r1 )  R(r1 , r2 )  . . . in R is strictly increasing and so must have finite length. A regular sequence is maximal if it cannot be extended to a longer regular sequence. Regular sequences do not necessarily remain regular when permuted, see Kaplansky [59, pg. 102], but the situation is nicer for homogeneous regular sequences. Proposition 2.7.1. Let R be a finitely generated commutative algebra graded over K of any characteristic p ≥ 0. Then any permutation of a homogeneous regular sequence z1 , z2 , . . . , zk is again a (homogeneous) regular system. Proof. Consider the case where w, z is a regular sequence, that is, suppose that w is not a zero-divisor in R, that z is not a zero-divisor in R/(w) and suppose that rz = 0 for some r in R. We note that since z is homogeneous, we may assume that r is homogeneous of minimal degree with this property. However, since z is not a zero-divisor modulo w, we have that r = sw for some homogeneous element s ∈ R. But then rz = swz = 0, which implies sz = 0 contradicting the minimal degree property of r. Therefore, z cannot be a zerodivisor in R. Now suppose the image w of w in R/(z) is a zero-divisor. Then there are homogeneous elements q, r ∈ R with rw = qz. But z is not a zerodivisor modulo (w), so it must be that q = tw for some homogeneous t ∈ R, and then rw = twz, so r = tz, since w is not a zero-divisor in R, and therefore, r = 0 as required.

32

2 Elements of Algebraic Geometry and Commutative Algebra

In the general case, because the group of permutations is generated by transpositions of the form, (,  + 1), it is sufficient to show that z1 , . . . , z−1 , z+1 , z , . . . , zk is regular. Now z1 , . . . , z−1 is regular, so it suffices to show that z+1 , z , . . . , zn is regular in R = R/(z1 , . . . , z−1 ). But z+2 , . . . , zn is regular in R = R/(z1 , z2 , . . . , z , z+1 ), so the result follows from the previous paragraph.  

2.8 Cohen-Macaulay Rings The depth(R) of a ring local ring (R, m) is defined as the maximal length of a regular sequence in m. It can be shown that in a graded Noetherian ring R any two homogeneous maximal regular sequences have the same length. If R is a graded Noetherian ring we put depth(R) = depth(R+ ). The depth of a ring is always less than or equal to its Krull dimension. A local Noetherian ring (R, m) is said to be Cohen-Macaulay if depth(m) = height(m) and a general Noetherian ring R is said to be Cohen-Macaulay if the localization of R at each of its prime ideals is Cohen-Macaulay. It can be shown (see for example [6, Section 4.3]) that a Noetherian graded algebra A is Cohen-Macaulay if and only if there is some homogeneous system of parameters for A which is also a regular sequence. In our setting it is often easier to use this condition to show that a ring is Cohen-Macaulay. For example, it is easy to show that x1 , x2 , . . . , xn is a regular sequence in K[x1 , x2 , . . . , xn ] which shows that K[x1 , x2 , . . . , xn ] is a Cohen-Macaulay ring. Cohen-Macaulay rings enjoy a number of useful properties. For nonmodular representations of finite groups, a theorem due to Hochster and Eagon [53] shows that the ring of invariants is always Cohen-Macaulay. We will show in Chapter 9, Theorem 9.2.2, that very few modular representations have a Cohen-Macaulay ring of invariants. This is one of the main reasons why modular rings of invariants are more difficult to handle. We now give a fundamental theorem of Noetherian algebra. The usual proofs of this theorem use homological algebra, e.g. Smith [102, Corollary 6.7.7]). We give a (previously unpublished) proof of Ian Hughes which does not use any homological algebra. In our opinion, Hughes proof is more straightforward than any of the other proofs of which we are aware. Theorem 2.8.1. Let A be a finitely generated connected graded K-algebra which is Cohen-Macaulay. Then every homogeneous system of parameters for A is a regular sequence for A. Proof. The proof is by induction on n = dim A. If n = 0 there is nothing to prove. Suppose n = 1. This implies that there exists a homogeneous element u ∈ A with u not a zero divisor (and u ∈ / A0 ). We let y be a homogeneous

2.8 Cohen-Macaulay Rings

33

system of parameters for A. We must show that y is not a zero divisor. Now u satisfies some monic polynomial of degree r (say) with coefficients in K[y]. Since both u and y are homogeneous, this implies that if a ∈ A with ay = 0, then aur = 0. But u is not a zero divisor and so a = 0. This shows that y is not a zero divisor as required. We now suppose that n > 1. Let y1 , y2 , . . . . , yn be a homogeneous system of parameters for A. We first show that A has a homogeneous system of parameters which is a regular sequence for A which is contained in K[y1 , y2 , . . . , yn ]. Since A is Cohen-Macaulay, it has some homogeneous system of parameters which is a regular sequence. Let u denote the first element in this regular sequence and for x ∈ A write x to denote the image of x in A/(u). Now A/(u) is Cohen-Macaulay with dim A/(u) = n − 1. Since A is integral over K[y1 , y2 , . . . , yn ], we know A/(u) is integral over K[y1 , y2 , . . . , yn ]. Therefore, by the Noether Normalization Lemma, there are homogeneous elements u2 , u3 , . . . , un ∈ K[y1 , y2 , . . . , yn ] such that u2 , u3 , . . . , un is a homogeneous system of parameters for A/(u). By induction, this is a regular sequence for A/(u) and so u, u2 , u3 , . . . , un is a homogeneous system of parameters for A which is a regular sequence for A. But then u2 , u, u3 , . . . , un is a regular sequence for A by Proposition 2.7.1. Note that u2 ∈ K[y1 , y2 , . . . , yn ]. We repeat the above argument using u2 in the role of u and conclude that A has homogeneous system of parameters z1 = u2 , z2 , . . . , zn contained in K[y1 , y2 , . . . , yn ] which is a regular sequence for A. Now y1 is integral over K[z1 , z2 , . . . , zn ] and so y1 satisfies a monic polynomial of degree r (say) with coefficients in K[z1 , z2 , . . . , zn ] ⊆ K[y1 , y2 , . . . , yn ]. r−1 Thus y1r = − i=0 ai y1i with ai ∈ K[z1 , z2 , . . . , zn ]. Expressing each ai as a polynomial in y1 , y2 , . . . , yn we see that one of the terms ai includes the monomial y1r−i . From this, using the homogeneity of the yi and of the zi , it follows that there exists s−1some k, and some non-zero scalar λ for which we have λzk = y1s + i=0 bi y1i with homogeneous bi ∈ K[y2 , . . . , yn ] for i = 0, 1, . . . , s − 1. For x ∈ A, we now denote by x the image of x in A/(zk ). Then the above implies that y1 is integral over K[y2 , y3 , . . . , yn ]. Thus A/(zk ) is integral over K[y2 , y3 , . . . , yn ] and so, since dim A/(zk ) = n − 1, we have that y2 , y3 , . . . , yn is a homogeneous system of parameters for A/(zk ). So, by induction, it is a regular sequence for A/(zk ). This implies that zk , y2 , . . . , yn is a regular sequence for A. By Proposition 2.7.1, y2 , y3 , . . . , yn , zk is also a regular sequence for A. But zk = λ−1 y1s + b with b in the ideal A(y2 , y3 , . . . yn ) of A. Thus y2 , y3 , . . . , yn , y1s is a regular sequence for A which implies that y1 , y2 , . . . , yn is also a regular sequence for A.   The following theorem (see [6, Theorem 4.3.5] for a proof) gives two more important characterizations of Cohen-Macaulay algebras. Theorem 2.8.2. Let A be a graded connected Noetherian K-algebra. The following are all equivalent. 1. A is a Cohen-Macaulay ring of Krull dimension n.

34

2 Elements of Algebraic Geometry and Commutative Algebra

2. There exists a homogeneous system of parameters y1 , y2 , . . . , yn in A such that A is a free K[y1 , y2 , . . . , yn ]-module. 3. A is a free K[y1 , y2 , . . . , yn ]-module for every homogeneous system of parameters y1 , y2 , . . . , yn in A.

2.9 The Hilbert Series Let R be a finitely generated graded K-algebra. The fact that R is finitely generated implies that each of its homogenous components Rd is a finite dimensional K vector space. We define the Hilbert series of R to be the power series  H(R, λ) = dimK (Ri )λi . i≥0

This series is sometimes called the Poincar´e series of R as well. For example, suppose R = K[h1 , . . . , hn ] is a polynomial algebra on generators of degrees di . Then H(R, λ) =

n  i=1

1 . (1 − λdi )

This is apparent when we consider that 1 = 1 + λd + λ2d + · · · + λmd + · · · . 1 − λd Suppose R is a graded Cohen-Macaulay ring and h1 , h2 , . . . , hn is a homogeneous system of parameters for R. Putting H = K[h1 , h2 , . . . , hn ] we have a Hironaka decomposition R = ⊕ri=1 Hfi on generators fi , of degrees mi for i = 1, . . . , r. Then λm1 + · · · + λmr . H(R, λ) = n di i=1 (1 − λ ) We are now in a position to define what it means for a Noetherian graded domain R over a field K to be Gorenstein. Namely we require that R is CohenMacaulay and that there exists an integer m such that H(R, λ−1 ) = λm (−1)dim R H(R, λ). Lemma 2.9.1. 1. A polynomial algebra R has a symmetric Hilbert series in the sense above, hence is Gorenstein. 2. A hypersurface has a symmetric Hilbert series in the sense above, hence is Gorenstein.  

2.10 Graded Nakayama Lemma

35

2.10 Graded Nakayama Lemma Let R be a finitely generated graded connected K-algebra. Let M be a nonnegatively graded R-module. An element f ∈ M is clearly not required as a generator of M if f lies in the submodule R+ M where R+ = ⊕∞ d=1 Rd . We let Q(M ) denote the (graded) vector space Q(M ) := M/R+ M . Here by a graded vector space we mean a vector space W which is a direct sum of vector spaces W = ⊕∞ d=0 Wd where elements of Wd are said to be homogenous of degree d. Proposition 2.10.1 (Graded Nakayama Lemma). Let R be a finitely generated graded connected K-algebra. Let M be a finitely generated nonnegatively graded R-module. Then the homogeneous elements f1 , f2 , . . . , fr generate M as an R-module if and only if their natural images f¯1 , f¯2 , . . . , f¯r span the K vector space Q(M ) = M/R+ M . Moreover, the elements f1 , f2 , . . . , fr minimally generate M if and only if {f¯1 , f¯2 , . . . , f¯r } is a vector space basis of Q(M ).  Proof. Suppose that {f¯1 , f¯2 , . . . , f¯r } spans Q(M ) and let N := ri=1 Rfi be the submodule generated by f1 , f2 , . . . , fr . Note that since the fi are homogeneous, N is a graded submodule of M . Assume by way of contradiction that N is a proper submodule of M and let d denote rthe least degree in which Nd  Md . Choose m ∈ Md \ Nd and write m ¯ = i=1 ci f¯i with ci ∈ K. Then r t m = i=1 ci fi + j=1 gj mj for some homogeneous gj ∈ R+ and mj ∈ M . By projecting this equation onto degree d we may assume deg(gj mj ) = d for i = 1, 2, . . . , t. Thus deg(mj ) < d for all j since deg gj ≥ 1. Therefore mj ∈ N for all j = 1, 2, . . . , t by the definition of d. This shows that m ∈ N , contradicting our assumption. Conversely it is clear that if f1 , f2 , . . . , fr generate M then their natural images span Q(M ). The final assertion of the Proposition follows immediately.   Note that any (homogeneous) lift of any homogeneous vector space basis for M/R+ M determines a minimal homogeneous generating set for M as an Rmodule. In particular, consider M = R+ . Let {f1 , f2 , . . . , fr } be homogeneous elements of R+ whose natural images {f¯1 , f¯2 , . . . , f¯r } form a vector space basis of Q(R+ ). Then f1 , f2 , . .  . , fr minimally generate R as a K algebra. To see r r this consider that R+ = i=1 Rfi . Thus if f ∈ Rd then f = i=1 gi fi for some homogeneous elements gi ∈ R+ . By induction on degree, we see that gi ∈ K[f1 , f2 , . . . , fr ] and thus f ∈ K[f1 , f2 , . . . , fr ]. Conversely, it is clear that 2 an element of R not lying in R+ cannot be written as a polynomial in lower degree elements of R and thus any homogenous algebra generating set must surject onto to a spanning set for Q(R+ ). In summary, we see that a homogeneous minimal generating set for R as an 2 K-algebra is obtained by lifting a homogeneous vector space basis of R+ /R+ . From this we see that such homogeneous minimal generating sets for R are

36

2 Elements of Algebraic Geometry and Commutative Algebra

not unique but that the number of generators of any given degree is fixed. We 2 ) is the embedding dimension of R. also note that dimF (R+ /R+ We apply the graded Nakayama lemma to a ring of invariants R = K[V ]G 2 to introduce terminology as follows. An invariant lying in R+ may be written as a sum of products of other invariants of lower degree. For this reason, an 2 is called decomposable. Conversely, an invariant (with invariant lying in R+ 2 is indecomposable and lies in some set of zero constant term) not lying in R+ G algebra generators for K[V ] .

2.11 Hilbert Syzygy Theorem Theorem 2.11.1 (Hilbert Syzygy Theorem). Let M be any finitely generated graded module over a polynomial ring A = K[x1 , x2 , . . . , xn ]. Then there is a finite graded resolution of M by free graded A-modules ρk

ρk−1

ρ2

ρ1

ρ0

0 −→ Mk −→ Mk−1 −→ · · · −→ M1 −→ M0 −→ M −→ 0 Moreover, we may take k ≤ n. Note that here we do not assume that deg(x ) = 1. We denote by I the kernel of the map ρ : M → M−1 . The theorem proceeds by choosing homogeneous generators for I−1 and a free module M mapping onto it. Therefore the maps at each stage depend upon this choice. If at each stage we choose a homogeneous minimal generating set for the kernel I of ρ then we obtain a homogeneous minimal free resolution. Although the maps depend upon our choice of ideal generators, the number k for which Ik is first zero does not, provided we choose a homogeneous minimal generating set for each ideal I , i.e., provided we work with a minimal free resolution. This number measures, in some sense, the complexity of M . It is called the projective dimension of M as an A-module and denoted pd(M ). We define the Betti numbers, β (M ), of M to be the rank of the th syzygy module: β (M ) := rankA (M ) when the free resolution of M is minimal. For example, β1 (M ) counts the number of minimal relations among any set of minimal homogeneous generators of M . Like the length of the resolution, k, the Betti numbers β (M ) do not depend upon the choices made in constructing a minimal free resolution of M . Using the grading on the M we may define the graded Betti numbers, βj (M ) := the number of minimal generators of M which have degree j. Again, the graded Betti numbers do not depend upon the choices made in constructing a minimal free resolution of M . Of course,  j≥0 βj = β for all . Thus pd(M ) = max{i | βi (M ) = 0}. Since the maps in the resolution preserve degree and since the resolution is everywhere exact, the Hilbert syzygy theorem yields H(M, λ) =

k  (−1)i H(Mi , λ). i=0

2.11 Hilbert Syzygy Theorem

37

n 1 Now we have H(A, λ) = =1 1−λdeg(x . Since each Mi is a free A-module, ) ri we may write Mi = ⊕j=1 Amij for some mi1 , mi2 , . . . , miri ∈ Mi . Therefore, n deg(x ) ). In H(Mi , λ) = (λdeg(mi1 ) + λdeg(mi2 ) + · · · + λdeg(miri ) )/  =1 (1 − λ  j terms of the graded Betti numbers this is H(Mi , λ) = ( j βij λ )/ n=1 (1 − λdeg(x ) ) and  i j ij (−1) βij λ . H(M, λ) = n deg(x ) ) =1 (1 − λ In particular, this shows that when A is a polynomial ring, the Hilbert series of any finitely generated graded A-module may be expressed as a rational function, i.e., as the quotient of two polynomials. By Noether’s Normalization Lemma, every Noetherian graded ring R is finitely generated over some polynomial ring A, and thus H(R, λ) is always a rational function of λ. Furthermore, any finitely generated R-module M is also a finitely generated A-module and thus H(M, λ) is a rational function of λ. An important example of the above for us will be when M = R is a Noetherian ring. For example, when R = K[V ]G . Let {f1 , . . . , fs } denote a minimal homogeneous generating set for R. Let A = K[x1 , x2 , . . . , xs ] denote the polynomial algebra on generators xi where we declare that deg(xi ) = deg(fi ) for all i = 1, 2, . . . , s. In our notation above, then, we have s = edim(R) = dim(A). Let ρ0 denote the surjective algebra map from A = M0 to R determined by ρ0 (xi ) = fi for i = 1, 2, . . . , s. The map ρ0 provides R with the structure of a finitely generated A-module and thus we may consider its resolution by syzygies. This turns out to be an effective way to study R and in particular to compute H(R, λ). In Chapter 13 we will study this technique in detail.

3 Applications of Commutative Algebra to Invariant Theory

In this chapter, we use some of the basic commutative algebra discussed in the previous chapter to develop some basic results about rings of invariants. Let R be a Noetherian graded K-algebra. Suppose that G is a group of degree-preserving automorphisms of R. Then G acts on Rd and we denote the elements of Rd fixed by G by RdG and we define RG = ⊕d≥0 RiG , so that RG is a graded connected algebra. If G acts on R as degree-preserving automorphisms, then it is easy to see that G also acts on Quot(R) as a group of automorphisms of the field. We may therefore form the invariant subfield Quot(R)G . Lemma 3.0.1. For any finite group G, we have Quot(R)G = Quot(RG ). Consequently, the extension Quot(R)G ⊂ Quot(R) is Galois, with group G and so Quot(R) has dimension |G| as a Quot(R)G vector space. Proof. If r/s ∈ Quot(R)G with r, s ∈ R, then  r · e=σ∈G σ(s) r := . s NG (s) Since the fraction itself and the denominator are both invariant, the numerator is also, finishing the proof.   Proposition 3.0.2. If R is a unique factorization domain, then R is integrally closed. In particular, F[V ] is integrally closed. Proof. Suppose that r/s ∈ Quot(RG ) is integral over R. We suppose that r and s do not share any common factor. There exist a0 , a1 , . . . , at−1 ∈ R such that (r/s)t + at−1 (r/s)t−1 + · · · + a1 (r/s) + a0 = 0. Therefore rt = −s(at−1 rt−1 + at−2 r t−2 s + · · · + a1 rst−2 + a0 st−1 ). Thus every irreducible factor of s divides r t and so must also divide r. But since r and s share no common factors, this implies that s must be a unit. Thus r/s ∈ R.   Proposition 3.0.3. Let R be an integrally closed domain on which the finite group G acts. Then RG is also integrally closed. H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 3, © Springer-Verlag Berlin Heidelberg 2011

39

40

3 Applications of Commutative Algebra to Invariant Theory

Proof. Suppose r/s ∈ Quot(RG ) is integral over RG . Then r/s is integral over R and lies in Quot(R)G ⊆ Quot(R). Since R is integrally closed, this implies that r/s ∈ R and since r/s is G-invariant, r/s ∈ RG .   Proposition 3.0.4. Let G be any finite group acting on a finitely generated algebra R as a group of automorphisms. Then R is integral over RG . In particular, R is a finitely generated RG -module. Proof. Suppose R is generated by elements {r1 , . . . , rs }. For t an indeterminate, consider the polynomials Fi (t) ∈ R[t] defined as Fi (t) =

 σ∈G

(t − σ(ri )) =

|G| 

ai,j ti .

j=0

We extend the action of G to automorphisms of R[t] by defining σ(t) = t for all σ ∈ G. With this convention we see from the definition of Fi (t) that the polynomials Fi (t) are invariant. Thus each coefficient ai,j on the right-hand side is invariant. Furthermore, we have shown that each of the generators of R is integral over RG , and therefore R is integral over RG as claimed. The final assertion of the proposition follows using Proposition 2.5.1.   Lemma 3.0.5. Using the notation of the above proof, {r1i1 · · · rsis | 0 ≤ ij ≤ |G| − 1} spans R as a module over RG .

 

The following important Corollary of Proposition 3.0.4 is immediate. Corollary 3.0.6. Let G be a finite group. Let f1 , f2 , . . . , fn ∈ F[V ]G be a sequence of homogeneous invariants. Then f1 , f2 , . . . , fn is a homogenous system of parameters for F[V ]G if and only if it is a homogenous system of parameters for F[V ]. Remark 3.0.7. Proposition 3.0.4 implies that if G is a finite subgroup of GL(V ) then dim K[V ]G = dim K[V ] = dimK V . It follows that the Krull dimension of RG is the same as the Krull dimension of R.

3.1 Homogeneous Systems of Parameters There are some especially useful homogeneous systems of parameters for F[V ]G . If our group is a permutation group, then the elementary symmetric functions form a homogeneous system of parameters, see Section 3.2. If F is finite then we may always use the Dickson invariants as a homogeneous

3.1 Homogeneous Systems of Parameters

41

system of parameters, see Section 3.3. If our representation is lower triangular, then M´ ui has constructed a homogeneous system of parameters, see Section 3.4. In particular, M´ ui’s result applies for any modular representation of a p-group (by Proposition 4.0.2). If the field F is infinite, then we can construct a homogeneous system of parameters by a method due to Dade, see [104]. First, we show that there must exist a basis {x1 , . . . , xn } for V ∗ which has the property that, for any ntuple σ1 , . . . , σn of elements of G, the set of linear forms {σ1 (x1 ), . . . , σn (xn )} is linearly independent. We begin by choosing any non-zero element x1 ∈ V ∗ . We choose the remaining basis elements x2 , x3 , . . . , xn inductively as follows. Having chosen x1 , x2 , . . . , xi we consider the union of proper subspaces of V ∗ defined by  Ki := spanF {σ1 (x1 ), σ2 (x2 ), , . . . , σi (xi )} . σ1 ,σ2 ,...,σi ∈G

construction depends upon the observation that, as a finite union of proper subspaces, Ki , cannot be all of V ∗ when F is infinite. We note also that Ki is stable under the action of G by definition. While i < n we may choose a linear form xi+1 ∈ V ∗ \ Ki . Such an element xi + 1 has the property that σ(xi+1 ) ∈ V ∗ \ Ki for all σ ∈ G. Now form the polynomials fi = NG Gx (xi ), i

for i = 1, 2, . . . , n. Each fi is of degree less than or equal |G|. To see that f1 , f2 , . . . , fn is a homogeneous system of parameters, suppose fi (v) = 0 for all i = 1, 2, . . . , n. Since each fi is a product of linear forms, there must exist, for each i, a group element σi ∈ G with σi (xi )(v) = 0. But the set {σ1 (x1 ), σ2 (x2 ), . . . , σn (xn )} is linearly independent in V ∗ and hence it is lin∗ early independent also in V = F ⊗ V ∗ , and hence v = 0, as required. Remark 3.1.1. If we are working over a finite field F we may work in the  of F to use Dade’s construction to construct a homogeneous algebraic closure F  ]G . Note that the homogeneous system of parameters, f1 , f2 , . . . , fn in F[V  G system of parameters will lie in F [V ] for a finite overfield F ⊃ F containing the finitely many coefficients of the fi . Probably the most famous result of Invariant Theory is David Hilbert’s proof [52] that (in modern terminology) the ring of invariants of a (geometrically) reductive group is finitely generated. However, his proof does not apply to modular representations since these fail to be linearly reductive. For a good discussion of these issues, see Derksen and Kemper’s book [26, section 2.2]. However, Emmy Noether [88] proved the following result. Theorem 3.1.2. The ring of invariants F[V ]G of a finite group is finitely generated.

42

3 Applications of Commutative Algebra to Invariant Theory

Proof. By the graded Nakayama lemma, Lemma 2.10.1, we need to prove 2 that Q(F[V ]G ) = F[V ]G /(F[V ]G + ) is a finite dimensional F vector space. Let F be the algebraic closure of F and consider V := V ⊗F F. It is clear that Q(F[V ]G ) ∼ = Q(F[V ]G ) ⊗F F. Therefore, it suffices to show that Q(F[V ]G ) is a finite dimensional F vector space. Hence, replacing F by F if necessary we may assume that F is infinite. Therefore, by Dade’s algorithm, there exists a homogeneous system of parameters f1 , f2 , . . . , fn for F[V ]G . By Corollary 3.0.6, these elements f1 , f2 , . . . , fn also form a homogeneous system of parameters for F[V ]. Let A denote the polynomial ring A = F[f1 , f2 , . . . , fn ]. Since F[V ] is a Cohen-Macaulay ring, there exists h1 , h2 , . . . , ht ∈ F[V ] such that F[V ] ∼ = ⊕ti=1 Ahi . Since F[V ] is a finitely generated A-module and A is a Noetherian ring, we see that F[V ] is a Noetherian A-module. It is clear that F[V ]G is an A-module and thus as an A-submodule of a Noetherian A-module, F[V ]G is also a Noetherian A-module. r Thus there exists a finite set g1 , g2 , . . . , gr ∈ F[V ]G with F[V ]G = j=1 Agj . Therefore   F[V ]G = F[f1 , f2 , . . . , fn , g1 , g2 , . . . , gr ] is a finitely generated F-algebra. In the notation of the above proof, the elements f1 , f2 , . . . , fn are traditionally called primary invariants and the elements g1 , g2 , . . . , gr secondary invariants. It is important to notice that there are infinitely many choices for both the primary and secondary invariants. These two terms merely describe the role of certain invariants within a particular fixed generating set. Since F[V ]G is finitely generated, its spectrum, Spec(F[V ]G ) is an affine variety which we denote by V //G . Dual to the natural inclusion of algebras is the natural surjection of varieties πV,G : V → V //G. −1 (πV,G (v)) = Gv. The geThe fibres of this surjection consist of G-orbits: πV,G ometry and topology of V //G reveals information about the ring of invariants, F[V ]G , and vice versa.

Proposition 3.1.3. [105, Proposition 2.2.3] Let {f1 , f2 , . . . , fn } be any homogeneous system of parameters for the polynomial ring S = F[x1 , x2 , . . . , xn ] and suppose nthe degree of fi is di . Then S is a free R = F[f1 , f2 , . . . , fn ]module on i=1 di many generators. Proof. Since S is Cohen-Macaulay, we have a Hironaka decomposition S = ⊕m i=1 Rηi for some homogeneous η1 , η2 , . . . , ηm ∈ S. Using this decomposition we see that the Hilbert series of S is given by

3.1 Homogeneous Systems of Parameters

H(S, λ) = =

Since H(S, λ) =

n

i=1 (1

m  j=1

H(R, λ)λdeg(ηj )

j=1 m deg(ηj ) λ nj=1 di i=1 (1 − λ )

− λ)−1 we see that

deg(ηj )

λ

m 

43

n (1 − λdi ) = i=1 n i=1 (1 − λ) =

n 

(1 + λ + λ2 + . . .+ λdi −1 )

i=1

Setting λ = 1 here gives m =

n i=1

di .

 

Corollary 3.1.4. [26, Theorem 3.7.1] Let {f1 , f2 , . . . , fn } be any homogeG G neous system of parameters of degrees d1 , d2 , . . . , dn for nF[V ] . Then F[V ] is a free R = F[f1 , . . . , fn ]-module if and only if it has ( i=1 di )/|G| many generators as an R-module. If F[V ]G is not free as an R-module, more generators are required. Proof. Using Proposition 3.1.3 we have F[V ] = ⊕m i=1 Rηi n where m = i=1 di . Hence {η1 , η2 , . . . , ηm } is a vector space basis for F(V ) as an Quot(R)-vector n space. Thus the degree of the field extension F(V ) over Quot(R) is i=1 di . By Lemma 3.0.1, the field extension F(V ) over Quot(F[V ]G ) has degree |G|. Therefore, the degree of the subfield extenn G ) over Quot(R) is ( d )/|G|. Thus F(V )G has  dimension sion Quot(F[V ] i i=1 n ( i=1 di )/|G| as a F(f1 , f2 , . . . , fn )-vector space. Suppose F[V ]G = sj=1 Rhj G where s is minimal. Then n {h1 , h2 , . . . , hs } spans F(V ) as a Quot(R)-vector space and  thus s ≥ ( i=1 di )/|G|. If s > ( ni=1 di )/|G|, then {h1 , h2 , . . . ,  hs } ⊂ F(V )G is an Quot(R) linearly a s dependent set. Therefore, we may write j=1 bjj hj = 0 for some aj , bj ∈ R s with at least one aj = 0. Clearing denominators gives j=1 aj hj = 0 with a1 , a2 , . . . , as ∈ R not all zero, showing that F[V ]G is not a free R-module   (and thus that F[V ]G is not Cohen-Macaulay). G Remark 3.1.5. In the above nproof we showed that the field extension F(V ) over Quot(R) has degree ( i=1 di )/|G|. This shows that if {f1 , . . . , fn } is any G homogeneous n system of parameters of degrees d1 , . . . , dn for F[V ] , then |G| divides i=1 di .

44

3 Applications of Commutative Algebra to Invariant Theory

Corollary 3.1.6. Let {f1 , f2 , . . . , fn } be any homogeneous system of parameters of degrees d1 , d2 , . . . , dn for F[V ]G . Then n 

di = |G| if and only if F[V ]G = F[f1 , f2 , . . . , fn ] .

i=1

3.2 Symmetric Functions Consider G = Σn ⊂ Gl(V ) acting as all permutations of a basis n{x1 , . . . , xn } ∗ . We note that Gx = {x , . . . , x } and that H (t) = for V 1 1 n x 1 i=1 (t − xi ) = n j n−j (−1) s t . Here, the s is the j-th elementary symmetric function j j j=0 sj =



xi1 xi2 · · · xij .

1≤i1 m.  Since deg(f ) > m, we see that f ∈ (K[V ]+ )m ⊆ I and we may write f = ri=1 ki hi where each ki is a homogeneous element of K[V ]+ . Since |G| is invertible, we may average over the orbit to obtain 1  1  1  f= σj (f ) = σj (ki hi ) = σj (ki )hi |G| j=1 |G| j=1 i=1 |G| j=1 i=1 m

r

m

r

1  σj (ki )hi . |G| r

=

m

m

i=1 j=1

m

Since j=1 σj (ki ) ∈ K[V ]G + , this shows that f is a decomposable invariant  and so cannot be part of any minimal algebra generating set for K[V ]G .  The above proof shows that in the non-modular case, the Hilbert ideal is generated by homogeneous elements of degree at most |G|. Kemper [26, Conjecture 3.8.6 (b)] has made the following conjecture.

48

3 Applications of Commutative Algebra to Invariant Theory

Conjecture 3.5.2 (Kemper). Let V be a representation of a finite group G. The Hilbert ideal (K[V ]G + )K[V ] is generated by homogeneous elements of degree at most |G|.

3.6 Representations of Modular Groups and Noether’s Bound In this section, we describe work of Karagueuzian and Symonds [62] and Symonds [106]. Theorem 3.6.1. If F is finite of characteristic p and G is a finite group, then F[V ] has only finitely many isomorphism types of indecomposable summands. The theorem has real force when G is modular, for then G has, in general, a “wild” representation theory: it is not possible to classify the indecomposable representations of G. Example 3.6.2. We recall the following material from the paper of Elmer and Fleischmann [36], where a much more complete discussion occurs. Let G = C2 × C2 be generated by elements σ, τ . It is known that this group has countably many isomorphism classes of indecomposable representations. Consider the 3-dimensional representation, V , of G over a field, F, of characteristic 2 given by ⎞ ⎞ ⎛ ⎛ 101 100 ρ(σ) = ⎝0 1 1⎠ and ρ(τ ) = ⎝0 1 0⎠ . 001 001 We let U denote the trivial 1-dimensional representation of G and we let W denote the regular 4-dimensional representation of G. Then the symmetric algebra of V over F contains the indecomposable summands U , V and W and no others. Recalling the notation and result of section §3.4, we saw that the invariants of the upper triangular group G = Un (Fq ) are generated by the orbit polynomials  (xi + v), where Vi−1 is the span of {x1 , . . . , xi−1 } . hi = v∈Vi−1

That is, F[x1 , x2 , . . . , xn ]G = F[h1 , h2 , . . . , hn ]. Let I denote the integer sequence (1, 2, . . . , n) and let I  = (1, 2, . . . , n − 1), and, for any subsequence J ⊂ I  , we define J c = I \ J. Finally, we define H(J) = F[h |  ∈ J c ]. We note that G acts trivially on H(J) for any J. Given any p-group P , we may embed P in G (see §4.0.2), and hence H(J) is P -invariant under this embedding for any J. We define finite dimensional graded FP -modules X(P, J) ⊂ F[V ]. Then

3.6 Representations of Modular Groups and Noether’s Bound

49

Theorem 3.6.3. There is an isomorphism of graded FP -modules F[V ] ∼ = ⊕J⊂I  H(J) ⊗F X(P, J). This structure theorem states that F[V ] contains one copy of X(P, J) for each monomial of H(J). The general case of a modular group follows. Remark 3.6.4. We note that there are 2n−1 − 1 such subsets, for any P . In [20], Campbell and Selick constructed 2n subsets M (i) of F[V ], each of which is an injective module over Steenrod’s algebra A. The module M (0) is the ring of invariants of the non-modular group F∗ . The relationship between the modules X(P, J) is not known to the authors of this book. Example 3.6.5. Let G denote the group U2 (F), so that F[V ]G = F[h1 , h2 ], where h1 = x1 and h2 = xq2 − xq−1 x2 . We set T = F[V ]/(h2 ). Then, as 1 modules over FG, we have F[V ] = F[h2 ] ⊗F T

where

T = X(G, {1}) ⊕ X(G, {∅}) .

Then, as modules over FG, we have F[V ] = F[h1 , h2 ] ⊗ X(G, ∅) ⊕ F[h2 ] ⊗F T (G, {1}) where X(G,∅) is the module spanned by {1} and T (G, {1}) is the module . The interested reader should compare to §7.2. spanned by x2 , x22 , . . . , xq−1 2 Example 3.6.6. Karagueuzian and Symonds [61, §2, extended example]: the case n = 3 and p = 3. Let G denote the group U3 (F), so that F[V ]G = F[h1 , h2 , h3 ] where |hi | = 3i−1 . We set T = F[V ]/(h3 ). Then, as modules over FG we have F[V ] = F[h1 , h2 , h3 ] ⊗ T (G, ∅) ⊕ F[h1 , h3 ] ⊗ X(G, {2}) ⊕ F[h2 , h3 ] ⊗ X(G, {1}) ⊕ F[h3 ] ⊗ X(G, {1, 2}) . Theorem 3.6.7. Suppose {f1 , f2 , . . . , fn } is a homogeneous system of parameters for the ring of invariants of the modular group G. Then F[V ]G is generated by module generators over F[f1 , f2 , . . . , fn ] of degrees less than or equal to n  (|fi | − 1) . i=1

Moreover, the relations among the module generators have degrees less than or equal to n  (|fi | − 1). 1+ i=1

Finally, the degree of P(G, V, t) as a rational function in t is at most −n.

50

3 Applications of Commutative Algebra to Invariant Theory

Corollary 3.6.8. [106] If F is finite and G is a non-trivial finite group acting on V with dimF (V ) > 1, then F[V ]G is generated in degrees less than or equal to dimF (V )(|G| − 1). These conclusions follow from the following theorem of Symonds, [106] Theorem 3.6.9. The invariant ring F[V ]G has Castelnuovo-Mumford regularity 0. Remark 3.6.10. Symonds proves more: if G acts on F[x1 , x2 , . . . , xn ] by homogeneous linear substitutions of the xi ’s, then the invariant ring has CastelnuovoMumford regularity 0.

3.7 Molien’s Theorem The following theorem of Molien, provides a constructive method to compute the Hilbert series of a ring of invariants of a finite group in characteristic 0. Theorem 3.7.1. (Molien) Let K be a field of characteristic zero. H(K[V ]G , λ) =

1  1 ( ). |G| det(IdV − λσ) σ∈G

Proof. Since the dimension of a vector space is unchanged under field extensions, we may assume that K is algebraically closed. Let W be any K representation of G. By the orthogonality relations for characters, see for example the book of Dummit and Foote [32][Chapter 18,  Theorems 15 and 16] or Lang [75][XVIII §5]), we have dimK (W G ) = 1 σ∈G trace(σ|W ). |G| Take σ ∈ G and consider the matrix A ∈ GL(V ) representing the action of σ on V . Considering the Jordan Normal form of A we see that since A has finite order and K is characteristic zero, the matrix A must be diagonalizable. Let ρ1 , ρ2 , . . . , ρn denote the n eigenvalues of A. Thus, working with a basis of eigenvectors, we may assume that A = diag(ρ1 , ρ2 , . . . , ρn ). Write R := K[V ]. Let x1 , x2 , . . . , xn be the dual basis of V ∗ . Thus the full set of monomials of degree m, {xa1 1 xa2 2 · · · xann | a1 + a2 + · · · + an = m} forms an eigenbasis for the action of A on Rm The trace of the action of σ on Rm is thus  1 −a2 n ρ−a ρ2 · · · ρ−a . trace(σ|Rm ) = n 1 a1 +a2 +···+an =m

Therefore,

3.7 Molien’s Theorem

51

1 1 = det(IdV − λσ) (1 − ρ1 λ)(1 − ρ2 λ) · · · (1 − ρn λ) n  = (1 + ρi λ + ρ2i λ2 + . . . ) =

=

i=1 ∞ 







ρa1 1 ρa2 2

m=0 a1 +a2 +···+an =m ∞    −1 ) λm trace(σ|R m m=0

· · · ρann

λm

.

Thus, H(RG , λ) =

∞ 

G (dimK Rm )λm

m=0  ∞ 

 1  = trace(σ|Rm ) λm |G| m=0 σ∈G 1 1  = |G| det(IdV − λσ −1 ) σ∈G 1 1  = . |G| det(IdV − λσ) σ∈G

  Remark 3.7.2. Suppose |G|−1 ∈ K, that is, G is a non-modular group. Elements of representation theory give us a complex representation of G, known as the Brauer lift, which shares the same Hilbert series as K[V ]G , see [25][§82]. Thus Molien’s Theorem can be used to compute the Hilbert series of the ring of invariants for any non-modular representation of a finite group. Remark 3.7.3. Note that if σ and τ are two elements of G lying in the same conjugacy class, then σ and τ have the same eigenvalues on V and therefore, 1 det(IdV − λσ)

=

1 det(IdV − λτ )

.

3.7.1 The Hilbert Series of the Regular Representation of the Klein Group Consider

⎛

σ = σ −1

0 ⎜1 =⎜ ⎝0 0

1 0 0 0

0 0 0 1

⎛ ⎞ 0 0 ⎜ 0⎟ ⎟ and τ = τ −1 = ⎜0 ⎝0 1⎠ 0 1

0 0 1 0

0 1 0 0

⎞ 1 0⎟ ⎟. 0⎠ 0

52

3 Applications of Commutative Algebra to Invariant Theory

The group G = {I4 , σ, τ, στ } is the Klein 4-group. We consider the ring of invariants K[V ]G where K is a field of any characteristic. Since G acts by permutations on V , the Hilbert series H(K[V ]G , t) is the same for all characteristics. Using Molien’s Theorem we compute   1 1 3 + H(K[V ]G , λ) = 4 (1 − λ)4 (1 − λ2 )2 3 1 = (1 + λ + λ2 + λ3 + λ4 + . . . )4 + (1 + λ2 + λ4 + λ6 + . . . )2 4 4 1 2 3 4 = (1 + 4λ + 10λ + 20λ + 35λ + . . . ) 4 3 + (1 + 2λ2 + 3λ4 + . . . ) 4 = 1 + λ + 4λ2 + 5λ3 + 11λ4 + . . . Note that H(K[V ] , λ) = G

= = =

  1 3 1 + 4 (1 − λ)4 (1 − λ2 )2 2 (1 + λ) + 3(1 − λ)2 4(1 − λ)4 (1 + λ)2 1 − λ + λ2 (1 − λ)4 (1 + λ)2 1 + λ3 . (1 − λ)(1 − λ2 )3

This is the Hilbert series we would expect to find if K[V ]G possessed a homogeneous system of parameters in degrees 1, 2, 2, 2 over which the invariants are a rank 2 free module with generators in degrees 0 and 3. Suppose there is such a homogeneous system of parameters. Looking in degrees 1 and 2 we see that (essentially) the only possibility for such a system is a, b1 , b2 , b3 where a = x1 +x2 +x3 +x4 , b1 = x1 x2 +x3 x4 , b2 = x1 x3 +x2 x4 and b3 = x1 x4 +x2 x3 . We use Lemma 2.6.3 to study this. Suppose v = (v1 , v2 , v3 , v4 ) is a point in V at which a, b1 , b2 , b3 all vanish, that is, v1 + v2 + v3 + v4 = 0 v1 v2 + v3 v4 = 0 v1 v3 + v2 v4 = 0 v1 v4 + v2 v3 = 0 If v1 = 0, then we have v2 v3 = v2 v4 = v3 v4 = 0 from which we conclude two of these three are also zero. But a(v) = 0 forces the fourth to also be zero. So we may assume by symmetry that none of the v = 0. Assume there is a point v where all 4 coordinates are non-zero. Then we have

3.7 Molien’s Theorem

53

v3 v 4 v1 v2 v 4 v3 = − v1 v2 v 3 v4 = − v1 v2 = −

It follows that v12 = v22 = v32 = v42 . Hence 0 = a(v)2 = 4v12 . In characteristic not 2 we get a contradiction and so can conclude that a, b1 , b2 , b3 is a homogeneous system of parameters. However in characteristic 2 we see that v = (1, 1, 1, 1) is a point where a, b1 , b2 , b3 all vanish and hence they are not a homogeneous system of parameters in this characteristic. However, the elementary symmetric functions form a homogeneous system of parameters in all characteristics. The corresponding form of the Hilbert series is 1 + 2λ2 + 2λ4 + λ6 H(K[V ]G , λ) = (1 − λ)(1 − λ2 )(1 − λ3 )(1 − λ4 ) As we will see later in §4.7, there is a set of module generators for K[V ]G over K[V ]Σ4 of degrees 0, 2, 2, 4, 4 and 6 valid in any characteristic. 3.7.2 The Hilbert Series of the Regular Representation of C4 Consider the regular representation of the cyclic group G := C4 with generator σ. The matrix of σ is given by ⎛ ⎞ 0001 ⎜1 0 0 0⎟ ⎟ σ=⎜ ⎝0 1 0 0⎠ 0010 We want to compute the ring of invariants K[V ]G where K is a field of any characteristic. As in the previous example, the Hilbert series H(K[V ]G , λ) is the same in all characteristics. Using Molien’s Theorem we compute   1 1 1 2 + + H(K[V ]G , λ) = 4 (1 − λ)4 (1 − λ2 )2 (1 − λ4 ) = 1 + λ + 3λ2 + 5λ3 + 10λ4 + 14λ5 + 22λ6 + . . . =

1 + λ2 + λ3 + 2λ4 + λ5 (1 − λ)(1 − λ2 )(1 − λ3 )(1 − λ4 )

This suggests that the ring of invariants K[V ]G is a free module over the ring of symmetric functions on generators of degrees 0, 2, 3, 4, 4, and 5. We will see in §4.8 that this is the case in every characteristic except 2. In fact, in characteristic 2, this ring of invariants is not even Cohen-Macaulay. Bertin [9] provided this ring of invariants as the first example of a unique factorization domain that was not Cohen-Macaulay answering a question of Pierre Samuel.

54

3 Applications of Commutative Algebra to Invariant Theory

3.8 Rings of Invariants of p-Groups Are Unique Factorization Domains Theorem 3.8.1. Let F be a field of characteristic p > 0 and let P be a pgroup. Then F[V ]P is a unique factorization domain. Proof. We proceed by induction on degree. If f ∈ F[V ]P has degree 1 then clearly, f is prime and there is nothing to prove. Suppose then, that f ∈ F[V ]P with deg(f ) > 1. Decompose f as a product of primes in F[V ]: f = f1 f2 · · · ft . Take σ ∈ P . Since σ(f ) = f we must have σ(f1 ) = λfj for some j with 1 ≤ j ≤ t and some λ ∈ F× . Without loss of generality, we may suppose that {1, 2, . . . , s} = {j | ∃ σ ∈ P, λ ∈ F∗ with σ(f1 ) = λfj }. Note that if σ ∈ P and 1 ≤ i ≤ s, then σ(fi ) = λfj for some j with 1 ≤ j ≤ s and some λ ∈ F∗ . Define h1 := f1 f2 · · · fs . Then for each σ ∈ P we must have σ(h1 ) = λ(σ)h1 for some λ(σ) ∈ F× . Now take σ, τ ∈ P . Then (στ )(h1 ) = σ(τ (h1 )) = σ(λ(τ )h1 ) = λ(τ )(σ(h1 )) = λ(τ )(λ(σ)h1 ) = λ(σ)λ(τ )h1 . Thus λ(στ ) = λ(σ)λ(τ ). This shows that λ : P → F× is a linear character of P . In particular, 1 = λ(e) = λ(σ p ) = λ(σ)p and therefore (1 − λ(σ))p = 0, whence λ(σ) = 1 for all σ ∈ P . This shows that h1 lies in F[V ]P . Moreover, for each j = 1, 2, . . . , s there exists σ ∈ P and λ ∈ F× such that σ(f1 ) = λfj . This shows that hi is an irreducible element in F[V ]P . Now by induction, the invariant f /h1 may be uniquely factored into a product of irreducibles in F[V ]P . Thus f /h1 = h2 h3 · · · hq where hj is irreducible in F[V ]P for 2 ≤ j ≤ q. Therefore, we have factored f = h1 h2 · · · hq into a product of irreducibles. Suppose now that f = h1 h2 · · · hr is another factorization of f into irreducibles in F[V ]P . Working in F[V ] we see that since the prime f1 of F[V ] divides f we must have that f1 divides hk for some k. Without loss of generality, we assume that k = 1. But since h1 ∈ F[V ]P , this implies that fj divides h1 for all j = 1, 2, . . . , s. Hence h1 divides the irreducible element h1 in F[V ]P and thus h1 and h1 are associated irreducible elements of F[V ]P . Write h1 = μh1 for some μ ∈ F∗ . Then f /h1 = h2 h3 . . . hq = μ−1 h2 h3 · · · hr . Now by induction, the element f /h1 of F[V ]P has a unique factorization up to ordering and scalar factors. Thus q = r and, renumbering if necessary, hj is an associate of hj for all j = 2, 3, . . . , q. Therefore, F[V ]P is a unique factorization domain.   Example 3.8.2. We take p = 3 and work over F3 . Let

3.9 When the Fixed Point Subspace Is Large

55

   a0 B= | a, d ∈ F3 \ {0}, b ∈ F3 bd be the set of all invertible 2×2 lower triangular matrices over F3 . Let V be the natural 2 dimensional representation of B. The p-Sylow subgroup U of B is the set of all the elements of B with 1’s along the main diagonal. Thus U is isomorphic to C3 , the cyclic group of order 3. We work with basis {e1 , e2 } of V with respect to which the elements of B are lower triangular and also the dual basis {x, y} of V ∗ . Thus e2 and x are eigenvectors for all the elements of B and fixed points for the elements of U . We have already seen in Theorem 1.11.2 that F3 [V ]U = F3 [x, N ] where N = N(y) = y 3 − x2 y. Thus F3 [V ]U is polynomial and so is clearly a unique factorization domain, as it must be by the theorem. In contrast, consider F3 [V ]B . Since U is a normal subgroup of B, we have F3 [V ]B = (F3 [V ]U )B/U = F3 [x, N ]{±1} . The non-trivial element −1 ∈ B/U acts on F3 [V ]U via −1 · x = −x and −1 · N = −N . Thus F3 [V ]B contains the irreducible elements x2 , xN and N 2 . Indeed, it is not too difficult to show that these 3 invariants generate F3 [V ]B . The two irreconcilable factorizations (xN ) · (xN ) = (x2 ) · (N 2 ) show that F3 [V ]B is not a unique factorization domain.

3.9 When the Fixed Point Subspace Is Large In this section, we consider the situation where the group fixes point-wise a subspace of codimension 1 or 2. In both cases, this guarantees that the ring of invariants is especially well-behaved. We will need the following lemma of J.P. Serre, his “Normality Criterion”. Recall that a Noetherian domain R is said to be integrally closed or normal, if R is integrally closed in its fraction field Quot(R). In other words, R is normal precisely if the following holds: if r, r ∈ R and there is a monic k condition i polynomial f (t) = i=0 ai t with coefficients ai ∈ R such that f ( rr ) = 0, then rr ∈ R. Theorem 3.9.1 (Serre’s Normality Criterion). [80, Theorem 23.8] A Noetherian domain R is integrally closed if and only if the following two conditions hold: (R1 ) (S2 )

If ℘ ∈ Spec(R) satisfies height(℘) ≤ 1, then R℘ is a regular local ring. If ℘ ∈ Spec(R) satisfies height(℘) ≥ 2, then depth(R℘ ) ≥ 2.

Part 1 of the following theorem was first proved by Landweber and Stong [73] with different techniques. Theorem 3.9.2. Let G be any subgroup of GL(V ) over any field K. Put n := dimK (V ). Then

56

3 Applications of Commutative Algebra to Invariant Theory

1. If dimK (V G ) = n − 1, then K[V ]G is a polynomial algebra. 2. If dimK (V G ) = n − 2, then K[V ]G is Cohen-Macaulay. Proof. We will prove both assertions simultaneously using a technique brought to our attention by Abraham Broer. First, we note that the question of whether K[V ]G is polynomial or is Cohen-Macaulay is unchanged under an extension of the underlying field K. For example, this can be seen from Corollaries 3.1.6 and 3.1.4. Thus we may assume that K is algebraically closed. For part 1, we define (V //G)good := {℘ ∈ Spec K[V ]G | K[V ]G ℘ is a regular local ring}. For part 2, we define (V //G)good := {℘ ∈ Spec K[V ]G | K[V ]G ℘ is a Cohen-Macaulay local ring}. In both cases, the set (V //G)good is an open subset of V //G. For part 1, this is shown in [71, VI Cor.1.16] and [80, Chapter 13]. For part 2, this is shown in [80, Chapter 8, Prop. (22.C)]. We also need the following two results. Let A be a finitely generated graded K-algebra with maximal homogeneous ideal m = A+ . 1. A is a polynomial ring if and only if Am is a regular local ring. 2. A is Cohen-Macaulay if and only if Am is Cohen-Macaulay. These statements are 2.2.25 and 2.1.27(c) of [14]. Let m denote the irrelevant ideal of K[V ]G , i.e., m = K[V ]G + . By the above two statements, we need to show that m ∈ (V //G)good . Let I denote the prime ideal I := I(V G ) ∈ Spec(K[V ]G ) and note that height(I) = codim(V G ). Now K[V ]G is integrally closed and hence by Theorem 3.9.1, it satisfies the two conditions (R1 ) and (S2 ). By (R1 ), K[V ]G I is a regular local ring if codim V G = 1. If codim V G = 2, then 2 = Krull dim(K[V ]G I ) and since the depth of any ring is at most its Krull dimension, we see that (S2 ) G G implies that 2 = Krull dim(K[V ]G I ) ≥ depth(K[V ]I ) ≥ 2 and thus K[V ]I is Cohen-Macaulay. Thus in both cases I ∈ (V //G)good . In particular, we see that (V //G)good is a non-empty open set. We claim that (V //G)good contains some maximal ideal n which contains I. To see this we define (V //G)bad := V //G \ (V //G)good . Let J denote the ideal J := I((V //G)bad ). Assume by way of contradiction that every maximal ideal n containing I lies in (V //G)bad , i.e., that every maximal ideal n containing I also contains J. Since I is a radical ideal, I is the intersection of the maximal ideals which contain it (see [80][Theorem 5.5], cf. the discussion in §2.2) and thus  n ⊇ J. I= n maximal n⊇I

Thus I ∈ I(J) = (V //G)bad . But this is a contradiction since we know I ∈ (V //G)good . This proves the claim.

3.9 When the Fixed Point Subspace Is Large

57

Thus there is a maximal ideal n containing I and with n ∈ (V //G)good . Let w denote the point of V G corresponding to the ideal n, i.e., {w} = V(n), and consider the map ψ : V → V defined by ψ(v) = v + w. Dual to this map we have the algebra homomorphism ψ ∗ : K[V ] → K[V ] given by ψ ∗ (f ) = f ◦ ψ. The map ψ ∗ is G-equivariant since for any σ ∈ G and any v ∈ V we have (σ(ψ∗ (f )))(v) = (ψ ∗ (f ))(σ −1 (v)) = (f ◦ ψ)(σ −1 (v)) = f (σ −1 (v) + w) = f (σ −1 (v + w)) = (σ(f ))(v + w) = (σ(f ) ◦ ψ)(v) = (ψ∗ (σ(f )))(v), i.e., σ(ψ ∗ (f )) = ψ ∗ (σ(f )). In particular, if f ∈ K[V ]G , then ψ ∗ (f ) ∈ K[V ]G . Thus restricting ψ ∗ gives an algebra automorphism of K[V ]G . Note that G ψ ∗ (n) = m. Therefore, ψ ∗ induces an isomorphism of K[V ]G n onto K[V ]m . Since n ∈ (V //G)good , this shows that m ∈ (V //G)good .  

4 Examples

In this chapter we collect a number of interesting examples and basic results in order to illustrate some of the techniques. As usual, our focus is two-fold: either determine a generating set of invariants, or determine the structure of the ring of invariants, or both. We begin with a useful lemma. Lemma 4.0.1. Suppose G is a p-group and F has characteristic p and let V be any positive dimensional representation of G. Then V G = {0}. Proof. By Lemma 1.10.3, there exists a composition series {e} = G0  G1  G2  . . .  Gm = G ∼ Cp for all i = 1, 2, . . . , m. with Gi /Gi−1 = We proceed by induction on the length of this composition series. The result is trivially true when m = 0 and G = {e}. Suppose by induction that W := V G = {0} and consider the action of G+1 /G on W . Let σ be a generator of the group G+1 /G ∼ = Cp . Consider the Jordan Normal Form for the action of σ on W . Choose a Jordan block with associated eigenvalue λ. Since σ has order p, we have λp = 1. But then (λ − 1)p = 0 and therefore, λ = 1. Therefore, the eigenvector associated to the block is in fact fixed by σ. Since G+1 /G is cyclic, this eigenvector is fixed by (G /G ) all of G+1 /G . Hence V G+1 = (V G )(G+1 /G ) = W +1  = {0}, finishing the proof.   Fixing a basis for V , we will denote by U (V ) the subgroup of GL(V ) consisting of lower triangular matrices with ones along the diagonal. We will also use the notation Un (F) to denote this subgroup when V has dimension n. Proposition 4.0.2. Let G be a p-group. Suppose V is an n dimensional representation of G defined over any field F of characteristic p. Then G is conjugate to a subgroup of U(V ). H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 4, © Springer-Verlag Berlin Heidelberg 2011

59

60

4 Examples

Proof. We will construct an ordered basis of V with respect to which every element of G is lower triangular. The proof proceeds by induction on the dimension n of V . If V has dimension 1, there is nothing to prove. In general, by the previous lemma, we note that W = V G has dimension at least 1. We consider the action of G on V /W . By induction, there is a basis for V /W with respect to which every element of G is lower triangular. Appending a basis for W to the chosen basis for V /W gives a lower triangular basis for V .   The following result, see Wilkerson [113], provides a very useful homogeneous system of parameters when working with a p-group. Proposition 4.0.3. Suppose that the action of G on V is lower triangular with respect to the basis {e1 , e2 , . . . , en } of V . Let {x1 , x2 , . . . , xn } be the dual G G basis of V ∗ . Then NG Gx1 (x1 ), NGx2 (x2 ), . . . , NGxn (xn ) is a homogeneous system of parameters for F[V ]G . Proof. The action of G on V ∗ is upper triangular. Thus for each i = 1, 2, . . . , n and each σ ∈ G, we may write σ(xi ) = ασ,i xi + qσ,i where ασ,i ∈ F× and  di qσ,i ∈ F[x1 , x2 , . . . , xi−1 ]. Therefore NG Gxi (xi ) = σ∈G/Gxi σ(xi ) = αi xi + qi for some αi ∈ F× and degxi (qi ) < di where di = [G : Gxi ]. Now we will use Lemma 2.6.3. To see that it applies, we must show that the only point where all n of the norms vanish is the point 0 ∈ V . To see this, consider v = G G (v1 , v2 , . . . , vn ) ∈ V(I) where I = (NG Gx (x1 ), NGx (x2 ), . . . , NGxn (xn ))F[V ]. 1

2

d1 Since q1 must be 0 we have NG Gx1 (x1 ) = α1 x1 and thus v1 = 0. Thus q2 (v) = G 0 and thus NGx (x2 )(v) = 0 forces v2 = 0. But then q3 (v) = 0 and so 2

NG (x3 )(v) = 0 implies v3 = 0. Continuing in this manner we see that v = (0, 0, . . . , 0). Thus by Lemma 2.6.3, we see that G G NG Gx (x1 ), NGx (x2 ), . . . , NGxn (xn ) 1

2

is homogeneous system of parameters for F[V ]G .

 

We are able to prove a more general result in Lemma 6.2.1 using term orders. The following example shows that F[3 V2 ]Cp is not Cohen-Macaulay. Example 4.0.4. Consider the representation V = 3 V2 of Cp over a field F of characteristic p. Fix a generator σ of Cp and choose an upper triangular basis {x1 , y1 , x2 , y2 , x3 , y3 } for (3 V2 )∗ with σ(yi ) = yi + xi and σ(xi ) = xi for i = 1, 2, 3. By Proposition 4.0.3, x1 , x2 , x3 , N(y1 ), N(y2 ), N(y3 ) is a homogeneous system of parameters for F[3 V2 ]Cp . However, x1 , x2 , x3 is not a regular sequence in F[3 V2 ]Cp . To see this, we use the elements uij = xi yj − xj yi for 1 ≤ i < j ≤ 3. These 3 elements are easily verified to be invariants. Furthermore, the relation x1 u23 − x2 u13 + x3 u12 = 0, which is easily verified, shows

4.1 The Cyclic Group of Order 2, the Regular Representation

61

that x3 · u12 = 0 in F[V ]Cp /(x1 , x2 ). Using deg(u12 ) = 2, it is easy to see that / F[V ]Cp (x1 , x2 ). Thus x3 is a zero divisor in F[V ]Cp /(x1 , x2 ). This shows u12 ∈ that x1 , x2 , x3 is a partial homogeneous system of parameters which is not a regular sequence in F[V ]Cp and hence that F[V ]Cp is not Cohen-Macaulay. A similar argument in a different setting is given in Example 9.1.6.

4.1 The Cyclic Group of Order 2, the Regular Representation Suppose G = C2 and V is a 2 dimensional K vector space. Let σ be a generator of G and let the action of G on V be given by   01 , σ = σ −1 = 10 with respect to a basis {e1 , e2 }. Let {x, y} denote the dual basis of V ∗ . Thus σ(xi y j ) = xj y i . In particular, (xy)i is invariant. If i = j then xi y j + xj y i is invariant. We claim that K[V ]G = K[x + y, xy]. Note that this representation is the special case n = 2 of the standard n dimensional representation of Σn . It is clear that K[V ]G ⊇ K[x + y, xy]. Assume, by way of contradiction, that K[x + y, xy] is a proper subset of K[V ]G . Let d be minimal such that K[x + y, xy]d  K[V ]G d. d i d−i Every element f ∈ K[V ]G d \ K[x + y, xy]d is of the form f = i=t ci x y where ct , ct+1 , . . . , cd ∈ K and ct = 0. Among all such f , fix one with t maximal. Define f  := f − ct (xy)t (x + y)d−2t . Then f  is invariant and f  = d  i d−i for some scalars ci . By the maximality of t, we must have i=t+1 ci x y  f ∈ K[x + y, xy]d . But then f = ct (xy)t (x + y)d−2t + f  ∈ K[x + y, xy]. This contradiction shows that K[V ]G = K[x + y, xy]. Of course, since C2 = Σ2 , the symmetric group on 2 letters, we recognize these two invariants as the elementary symmetric functions in x and y. This is the best possible situation in invariant theory, in which the ring of invariant polynomials is again polynomial algebra. We see from this calculation, or from Molien’s Theorem, that H(F[V ]G , λ) =

1 . (1 − λ)(1 − λ2 )

It is also worth noting that this calculation is independent of the field. Note that we could have calculated the Hilbert series in advance, and its form suggests that the ring of invariants is a polynomial ring on two invariants, one of degree 1 and the other of degree 2. The reader should be warned, however, that there are rings R for which the form of the Hilbert series resembles that of a polynomial ring even though R is no such thing, see Stanley’s paper [104, page 481].

62

4 Examples

4.2 A Diagonal Representation of C2 Suppose K is a field of characteristic different from 2. Then K contains a square root of 1, different from 1, namely -1. Suppose the generator, σ, of the cyclic group C2 acts on V ∗ via   −1 0 σ = σ −1 = 0 −1 with respect to the basis {x, y}. Then we note that σ(xi ) = (−1)i xi so that xi is invariant if and only if i = 2j. Similarly, y i is invariant if and only if i = 2j. Observe that x2j = (x2 )j and that σ(xi y i ) = xi y i is invariant. It isn’t hard to prove from here that K[V ]G = K[x2 , y2 , xy]. There are a variety of ways to consider this example; we demonstrate two particular viewpoints. First we have the isomorphism K[V ]G ∼ = K[a, b, c]/(c2 − ab) where deg(a) = deg(b) = deg(c) = 2. In another approach, we observe that {x2 , y 2 } forms a homogeneous system of parameters for K[V ]G , and that K[V ]G is a free module over K[x2 , y2 ] on the basis {1, xy}. In particular, we see that K[V ]G is a Cohen-Macaulay ring, for example by applying Corollary 3.1.4. The Hilbert series of this ring of invariants is H(K[V ]G , λ) =

1 − t4 . (1 − t2 )2

4.3 Fraction Fields of Invariants of p-Groups We recall here material from the paper of Campbell and Chuai [16]. Consider a representation V of a p-group G over a field F of characteristic p > 0. By Proposition 4.0.2, we can choose a basis {x1 , x2 , . . . , xn } for V ∗ such that (σ − 1)xi is in the span of {x1 , x2 , . . . , xi−1 } for all σ ∈ G and for all i = 1, 2, . . . , n. In particular, we note that x1 is invariant. We define R[m] := F[x1 , x2 , . . . , xm ] for 0 ≤ m ≤ n subject to the convention that R[0] = F. Then G acts on R[m]. For any non-zero f ∈ R[m], we may express f as a polynomial in xm and write f = f0 + f1 xm + · · · + fd xdm with fi ∈ R[m − 1] for all i = 0, 1, . . . , d and where fd = 0. The leading coefficient fd ∈ R[m − 1] of f plays a prominent role in our analysis and we therefore denote it by c(f ). Writing σ(xm ) = xm + αm−1 xm−1 + · · · +α1 x1 , we d have σ(f ) = i=0 σ(fi )(xm + αm−1 xm−1 + · · · + α1 x1 )i . Therefore, σ(c(f )) = c(σf ) for all σ ∈ G. In particular, if f is an invariant, so is c(f ).

4.3 Fraction Fields of Invariants of p-Groups

63

For each m with 1 ≤ m ≤ n, let φm ∈ R[m]G denote a fixed homogeneous invariant with the smallest positive degree in xm among all invariants in R[m]G . The existence of φm follows from the fact that the set  |G| R[m]G \R[m − 1] is non-empty since N (xm ) := σ∈G σ(xm ) = xm + {terms of lower degree in xm } lies in it. We take φ1 = x1 . The invariants cm = c(φm ) ∈ R[m − 1] will play a special role. Finally, note that we can make no claim as to the total degree of φm , in particular, we cannot claim that the total degree of φm is less than or equal to |G| for all m. If that were true, we would be able to prove that F(V )G is generated in degrees less than |G|, that is, that the Noether bound holds for the invariant fields of p-groups in characteristic p. We note, however, that Fleischmann, Kemper and Woodcock have proved that the Noether bound does holds for invariant fields for arbitrary representations of any finite group in any characteristic, see [40]. We first prove two lemmas. Lemma 4.3.1. For any f ∈ R[m]G , there exists an integer k ≥ 0 such that ckm f ∈ R[m − 1]G [φm ]. Proof. We use induction on degxm (f ). When degxm (f ) = 0, there is nothing to prove. So we may assume degxm (f ) = d > 0. In the ring R[m]cm , the element φm /cm is monic as a polynomial in xm . Hence we may divide f by φm /cm in order to obtain f = q  (φm /cm ) + r  where q  , r ∈ R[m]cm with degxm (r  ) < degxm (φm ). Thus f = σ(f ) = σ(q  )(φm /cm ) + σ(r  ) for all σ ∈ G. Since degxm (σ(r  )) = degxm (r  ) < degxm (φm ), we see by the uniqueness of remainders that r = σ(r ) and hence q  = σ(q  ) for all σ ∈ G. Therefore q , r ∈ R[m]G cm . Multiplying by a suitable power of cm , we  s  see that there exist an integer s ≥ 0 and polynomials q = cs+1 m q , r = cm r ∈ G s  R[m] such that cm f = qφm + r where degxm (r) = degxm (r ) < degxm (φm ). Therefore, r ∈ R[m − 1]G because φm has the least positive degree in xm inside R[m]G . Since degxm (q) = degxm (f ) − degxm (φm ), we see by induction that ctm q ∈ R[m − 1]G [φm ] for some t ≥ 0. Therefore, for k = s + t we have ckm f ∈ R[m − 1]G [φm ], as required.   We note that it follows immediately from Lemma 4.3.1 that if cm = 1 for all m = 1, 2, . . . , n, then any f ∈ R[m]G lies in F[φ1 , . . . , φm ] as easily seen by induction on m. We record this observation as Corollary 4.3.2. If cm = 1 for all m = 1, 2, . . . , n, then F[V ]G = F[φ1 , . . . , φn ] is a polynomial ring.

64

4 Examples

Lemma 4.3.3. For any finite number of invariants h1 , . . . , ht ∈ R[m]G , there exists a monomial c = ck11 · · · ckmm in c1 , . . . , cm , such that chi ∈ F[φ1 , . . . , φm ] for all i = 1, 2, . . . , t. Proof. We use induction on m. First let m = 1. Since φ1 = x1 and c1 = 1, the lemma follows from Corollary 4.3.2. Now assume m > 1. By Lemma G 4.3.1, there exists an integer s ≥ 0 such that csm h i ∈ R[m − 1] [φm ] for all s j i = 1, 2, . . . , t. For i = 1, 2, . . . , t, write cm hi = j=1id aij φm , where aij ∈ R[m − 1]G for all i and j. Now, since the finite set {aij | 1 ≤ i ≤ t, 1 ≤ j ≤ id } is contained in R[m − 1]G , we conclude by induction that there exist km−1 k1 , . . . , km−1 ≥ 0 such that ck1 1 · · · cm−1 aji ∈ F[φ1 , . . . , φm−1 ] for all i and km−1 s j. Hence we have, for c = ck11 · · · cm−1 cm , that chi ∈ F[φ1 , . . . , φm ] for all i = 1, 2, . . . , t, as required.   The following theorem shows that for a p-group, the invariant field is purely transcendental, a result originally due to Miyata [81]. This formulation, however, is constructive. Theorem 4.3.4. Let G ⊆ U(V ) ⊂ GL(V ) be a p-group. Choose any set of homogeneous invariants φ1 , . . . , φn with the property that φm ∈ R[m]G is of smallest positive degree in xm for 1 ≤ m ≤ n. Then F(V )G = F(φ1 , . . . , φn ). Furthermore, there exists a non-zero f ∈ F[φ1 , . . . , φn ] such that F[V ]G f = F[φ1 , . . . , φn ]f . Proof. We use the notation from above. For the first part of the theorem, we need only to show that any h ∈ F[V ]G lies in F(φ1 , . . . , φn ). Assume h ∈ R[m]G \R[m − 1]. By Lemma 4.3.1, there exists an integer s ≥ 0 such  that csm h ∈ R[m − 1]G [φm ]. We write csm h = dk=1 ak φkm , where ak ∈ R[m − 1]G for k = 1, 2, . . . , d. By Lemma 4.3.3, there exists some monomial cK = km−1 ck11 · · · cm−1 with cK · csm ∈ F[φ1 , . . . , φm−1 ] and cK · ak ∈ F[φ1 , . . . , φm−1 ] for all k = 1, 2, . . . , d. Thus h ∈ F(φ1 , . . . , φm ) ⊆ F(φ1 , . . . , φn ). For the proof of the second part, let F[V ]G = F[g1 , . . . , g ]. Then we can write gi = hi /f where hi ∈ F[φ1 , . . . , φn ] for i = 1, 2, . . . ,  and f ∈ F[φ1 , . . . , φn ] is non-zero. Then F[V ]G = F[h1 /f, h2 /f, . . . , h /f ] ⊆ F[φ1 , . . . , φn ]f , as required.

 

4.4 The Alternating Group We study An , the subgroup of Σn consisting of all even permutations acting via its usual n dimensional representation V . An alternating function is a polynomial f with σ(f ) = −f for all odd permutations σ. It is well known that a function invariant under An may be uniquely written as the sum of

4.5 Invariants of Permutation Groups

65

a symmetric function together with a symmetric function times the discriminant. Here the discriminant may be described as  Δn = (xj − xi ), 1≤i<j≤n

if p = 0 or if p > 2. When p = 2, we use the orbit sum xn−2 · · · xn−1 ). Δn := OAn (xn−1 1 2 We have the following Hironaka decomposition K[V ]An = K[V ]Σn ⊕ K[V ]Σn Δn . Hence, K[V ]An is generated by n + 1 elements as an algebra, and so K[V ]An is a hypersurface. It is easy to see that Δ2n is invariant under Σn when p = 0 or p > 2. To see this result in these latter cases, suppose p = 0 or p > 2, and let σ be any odd permutation so that Σn is generated by An together with σ. Suppose that f ∈ K[V ]An and define f+ := 12 (f + σ(f )) and f− := 12 (f − σ(f )). We note that for τ ∈ An , we have τ (f± ) = 12 (τ (f ) ± τ σ(f )) = 12 (f ± στ  (f )) = 1  2 (f ± σ(f )) = f± for some τ ∈ An because An is normal in Σn . Thus both f+ and f− are invariant under An . Clearly f+ is fixed by σ and hence is a symmetric function, while σ(f− ) = −f− . To finish the proof, we need to show that f− = f  Δn for some symmetric function f  . We observe that f− (x1 , x2 , . . . , xn ) = −f (x2 , x1 , . . . , xn ) since the transposition interchanging x1 and x2 is odd. Therefore, f− vanishes on the hyperplane determined by x2 −x1 = 0 and therefore, (x2 −x1 ) divides f− . Similarly, xj −xi divides f− for all 1 ≤ i < j ≤ n. Consequently, f− = f  Δn for some polynomial f  . Finally, for an odd permutation σ, we have σ(f− ) = σ(f  )σ(Δn ) = σ(f  )(−Δn ) and we conclude σ(f  ) = f  so that f  is, in  fact, symmetric.  If F = K has characteristic 2, then 1≤i<j≤n (xj −xi ) = 1≤i<j≤n (xj +xi ) is a symmetric function. However, as we noted above, we may define Δn := OAn (xn−1 xn−2 · · · xn−1 ) 1 2 which is not symmetric but is invariant under An . Lemma 4.5.3 states that with this definition for Δn , the usual Hironaka decomposition of F[V ]An as a module over F[V ]Σn is still valid in characteristic 2.

4.5 Invariants of Permutation Groups We say that V is a permutation representation of G if there exists a basis of V which is permuted by the action of G. We call this basis a permutation basis of V . In other words, V is a permutation representation of G

66

4 Examples

if G ⊂ Σn ⊂ GL(V ). A key observation is that when we work in the basis of V ∗ dual to the permutation basis, the elements of G permute the set of all monomials. Therefore, given a monomial xI , we form the orbit sum  I OG (x ) = σ∈G/H σ(xI ) where H = GxI is the isotropy group of xI . It is easy to see that OG (xI ) is a G-invariant polynomial. Lemma 4.5.1. The orbit sums OG (xI ) of degree d form a basis for K[V ]G d. Proof. Any f ∈ K[V ]d may be written as sum of monomials  a i xI . f= deg(I)=d

 But for any σ ∈ G we have σ(f ) = deg(I)=d aI σ(xJ ). It follows that if f is   G-invariant and xJ ∈ GxI then aJ = aI . The result is immediate. Corollary 4.5.2. Suppose that V is a permutation representation of G. The Hilbert series of K[V ]G depends only on G ⊂ Σn and not on the field K.   Lemma 4.5.3. The Hironaka decomposition F[V ]An = F[V ]Σn ⊕ F[V ]Σn Δn is valid over a field F of characteristic 2 with Δn = OAn (xn−1 xn−2 · · · xn−1 ). 1 2  

4.6 G¨ obel’s Theorem In this section we give a theorem of M. G¨obel [44] which provides a good generating set of orbit sums for the invariants of any permutation representation. In order to prove G¨ obel’s Theorem we introduce the concept of a gap. Consider a permutation representation V for a group G. Let {x1 , x2 , . . . , xn } be a permutation basis for V ∗ . Consider a monomial m = xa1 1 xa2 2 · · · xann = xA . We let set(A) denote the set of exponents set(A) := {a1 , a2 , . . . , an }. We define the height of xA , denoted ht(A) and ht(xA ), to be the largest exponent, ht(A) := max{ai | 1 ≤ i ≤ n}. We also define deg(A) := deg(xA ). We say the monomial xA has a gap or a gap at r if there exists a non-negative integer r such that {r + 1, r + 2, . . . , ht(A)} ⊆ set(A) but r ∈ / set(A). Note that xA does not have a gap means that set(A) = {0, 1, 2, . . . , ht(A)}. Theorem 4.6.1 (M. G¨ obel). Let V be a permutation representation of G. Then {OG (xA ) | xA does not have a gap} ∪ {x1 x2 · · · xn } is a generating set for K[V ]G .

4.6 G¨ obel’s Theorem

67

Proof. Given an exponent sequence A = (a1 , a2 , . . . , an ) we write Λs (A) := {j | aj = s} and λs (A) := |Λs (A)|. We define a partial order on exponent sequences (and the corresponding monomials) as follows. If deg(A) > deg(B) then we declare that A > B. When deg(A) = deg(B), we declare that A > B if there exists an integer t with λt (A) > λt (B) and λs (A) = λs (B) for all s > t. Note that if A and B are two sequences with A > B and B > A, then A and B must lie in the same Σn -orbit. If 0 ∈ / set(A) then xA and OG (xA ) are both divisible by the invariant / set(A) and A = (1, 1, . . . , 1), then sn = x1 x2 · · · xn . In particular, if 0 ∈ OG (xA ) is decomposable and so not required in any generating set. Thus we can and will assume from now on that 0 ∈ set(A). Suppose m = xA has gap at r. From m we define a new monomial m as follows:  if ai < r; ai , b1 b2 bn m = x1 x2 · · · xn where bi = ai − 1 if ai > r. Consider the product of orbit sums OG (m)OΣn (m/m). Note that OΣn (m/m) is the dth elementary symmetric function, sd , where d = deg(m/m). Then we can write  OG (m)OΣn (m/m) = cα OG (mα ) α∈X

where X is some index set, each mα is a monomial, each cα is a positive integer and where if α, β ∈ X are distinct, then OG (mα ) = OG (mβ ). Note that we can and will assume that each mα is of the form mα = m · ρ(m/m) for some ρ ∈ Σn . To see this, note that mα = σ(m) · τ (m/m) for some σ ∈ G and τ ∈ Σn . Since OG (mα ) = OG (σ −1 (mα )), we may replace mα by σ −1 (mα ) = m · σ −1 τ (m/m). In particular, we will assume that each mα is divisible by m. Consider one of the orbit sums OG (mα ) where α ∈ X. We claim that mα ≤ m in the partial order defined above. To see this, write m = xA and m = xB and mα = xC . Put t = ht(A). It is clear that ht(B) = t − 1 and that Λs (B) = Λs+1 (A) for s = r, r + 1, . . . , t − 1. Also, mα = m · τ (m/m) = (xb11 xb22 · · · xbnn ) · (xi1 xi2 · · · xid ) for some τ ∈ Σn and some 1 ≤ i1 < i2 · · · < id ≤ n. Taking Δ = {i1 , i2 , . . . , id } we have Λs (C) = (Λs (B) \ Δ)  (Λs−1 (B) ∩ Δ) = (Λs+1 (A) \ Δ)  (Λs (A) ∩ Δ) , for each s = r, r + 1, . . . , t.

68

4 Examples

Suppose that C ≥ A. Then λt (C) ≥ λt (A). Since Λt (C) = Λt (A) ∩ Δ, from the equation above, we must have Λt (C) = Λt (A) which implies that Λt (A) ⊆ Δ and λt (A) = λt (C). Hence λt−1 (C) ≥ λt−1 (A) because C ≥ A. But Λt−1 (C) = (Λt (A) \ Δ)  (Λt−1 (A) ∩ Δ) = Λt−1 (A) ∩ Δ. Therefore, Λt−1 (C) = Λt−1 (A), Λt−1 (C) ⊂ Δ and λt−1 (C) = λt−1 (A). Continuing in this manner we find that Λs (C) = Λs (A) and Λs (A) ⊂ Δ t for all s = r + 1, r + 2, . . . , t. But |Δ| = d = deg(m/m) = s=r+1 λs (A) which implies that Δ = ts=r+1 Λs (A). Therefore,  xi . τ (m/m) = i∈Λs

r+1≤s≤t

Thus our assumption that C ≥ A implies that C = A and that cα = 1. Therefore  cα OG (mα ) OG (m) = OG (m) · OΣn (m/m) − α∈Y

where |Y | = |X| − 1 and mα < m for all α ∈ Y . The fact that m has a gap implies that deg(m) < deg(m). Furthermore, deg(m) = 0 since m = sn since we have assumed that 0 ∈ set(A). Thus 1 ≤ deg(m) < deg(m). In summary, the assumption that m has a gap and m = sn implies that OG (m) may be expressed as a polynomial in orbit sums of monomials which are smaller than m in our partial order. Using induction with respect to the partial order this shows that if m has a gap and m = sn , then we do not require OG (m) as part of a minimal generating set of orbit sums for K[V ]G . Thus a generating set is formed by taking sn together with the orbit sums of all monomials without a gap.   Remark 4.6.2. Note that, in fact, the above proof proves the stronger statement that K[V ]G is generated as a module over K[V ]Σn by the set {OG (xI ) | I has no gaps}. Definition 4.6.3. An orbit sum OG (xA ) is called special if either A has no gaps or A = (1, 1, . . . , 1). Thus G¨ obel’s Theorem asserts that the ring of invariants of a permutation representation is always generated by its special orbit sums. Remark 4.6.4. If the permutation action of G on V acts transitively on the permutation basis of V , then the monomial sn is an indecomposable orbit sum. Conversely, if the permutation basis of V consists of r orbits with r ≥ 2, then sn is the product of r lower degree invariant monomials, each without a gap. Thus if G does not act transitively on the permutation basis of V , then the ring of invariants is generated by the orbit sums of the monomials without gaps. Corollary 4.6.5. Let V be a permutation representation of G with dim(V ) ≥ 3. Then

4.7 The Ring of Invariants of the Regular Representation of the Klein Group

69

  dim V . β(V, G) ≤ 2 Proof. The largest degree monomials without a gap are those monomials m n

. Since with set(m) = {0, 1, 2, . . . , n − 1}. Such monomials have degree 2

  n ≥ 3 we have n2 ≥ n = deg(sn ).

4.7 The Ring of Invariants of the Regular Representation of the Klein Group In this section and the next we examine in detail two examples in order to illustrate the use of G¨obel’s Theorem and Hilbert series. The first example is a continuation of the example we considered in §3.7.1. We are considering the regular representation of the Klein group G := C2 ×C2 over a field K of any characteristic. We choose a basis {x1 , x2 , x3 , x4 } for V ∗ with G-action given by x1 ↔1 x2

σ

x1 ↔2 x4

σ

x2 ↔2 x3 .

x3 ↔1 x4

σ σ

We begin by analyzing the structure of K[V ]G as a module over the ring K[V ]Σ4 since the elementary symmetric functions s1 , s2 , s3 , s4 form a homogeneous system of parameters for K[V ]G . By the proof of G¨ obel’s Theorem (see Remark 4.6.2), we know that the ring of invariants is generated as a module over K[s1 , s2 , s3 , s4 ] by the orbit sums without gaps. In addition to the elementary symmetric functions, the special orbit sums are associated to the decreasing exponent sequences (3, 2, 1, 0), (2, 2, 1, 0), (2, 1, 1, 0), (2, 1, 0, 0), For each such exponent sequence, as well as for the elementary symmetric functions, we need to write the Σ4 orbit sum as G orbit sums. In §3.7.1, we showed that H(K[V ]G , λ) =

1 + 2λ2 + 2λ4 + λ6 (1 − λ)(1 − λ2 )(1 − λ3 )(1 − λ4 )

The form of this Hilbert series means that if K[V ]G is Cohen-Macaulay, it must be generated as a module over K[s1 , s2 , s3 , s4 ] by 6 invariants of degrees 0, 2, 2, 4, 4 and 6. We note that these invariants may be chosen to be orbit sums by Lemma 4.5.1. Of course, if K[V ]G is not Cohen-Macaulay, then more generators will be needed, see Corollary 3.1.4. Writing out the Σ4 special orbit sums as G-orbit sums we get:

70

4 Examples

OΣ4 (x1 ) = s1 = OG (x1 ) OΣ4 (x1 x2 ) = s2 = OG (x1 x2 ) + OG (x1 x3 ) + OG (x1 x4 ) OΣ4 (x1 x2 x3 ) = s3 = OG (x1 x2 x3 ) OΣ4 (x1 x2 x3 x4 ) = s4 = OG (x1 x2 x3 x4 ) OΣ4 (x21 x2 ) = s1 s2 − 3s3 = OG (x21 x2 ) + OG (x21 x3 ) + OG (x21 x4 ) OΣ4 (x21 x2 x3 ) = s1 s3 − 4s4 = OG (x21 x2 x3 ) + OG (x21 x2 x4 ) + OG (x21 x3 x4 ) OΣ4 (x21 x22 x3 ) = s2 s3 − 3s1 s4 = OG (x21 x22 x3 ) + OG (x21 x2 x23 ) + OG (x21 x2 x24 ) OΣ4 (x31 x22 x3 ) = s1 s2 s3 − 3s21 s4 − 3s23 + 4s2 s4 = OG (x31 x22 x3 ) + OG (x31 x2 x23 ) + OG (x31 x2 x24 ) + OG (x31 x23 x4 ) + OG (x31 x22 x4 ) By G¨obel’s Theorem, the 20 G-orbit sums which occur on the right hand side of the above equations, together with 1, generate K[V ]G as a module over K[s1 , s2 , s3 , s4 ]. We will now show that the 6 orbit sums 1, OG (x1 x3 ), OG (x1 x4 ), OG (x21 x2 x4 ), OG (x21 x3 x4 ), OG (x31 x22 x3 ) suffice to generate K[V ]G as a module over K[s1 , s2 , s3 , s4 ]. Clearly, OG (x1 ) = s1 , OG (x1 x2 x3 ) = s3 and OG (x1 x2 x3 x4 ) = s4 are not required as module generators. The following identities, although tedious, are all easily verified. They show how the remaining 12 candidate orbit sums lie in the K[s1 , s2 , s3 , s4 ]-module generated by the 6 orbit sums listed above. OG (x1 x2 ) = s2 − OG (x1 x3 ) − OG (x1 x4 ) OG (x21 x2 ) = −s3 + s1 (s2 − OG (x1 x3 ) − OG (x1 x4 )) OG (x21 x3 ) = −s3 + s1 OG (x1 x3 ) OG (x21 x4 ) = −s3 + s1 OG (x1 x4 ) OG (x21 x2 x3 ) = (s1 s3 − 4s4 ) − OG (x21 x2 x4 ) − OG (x21 x3 x4 ) OG (x21 x22 x3 ) = −(s1 s4 − s2 s3 ) + s3 OG (x1 x4 ) + s3 OG (x1 x3 ) OG (x21 x2 x23 ) = −(s1 s4 ) + s3 OG (x1 x3 ) OG (x21 x2 x24 ) = −(s1 s4 ) + s3 OG (x1 x4 ) OG (x31 x22 x4 ) = −(s21 s4 − s1 s2 s3 + s23 ) − (s1 s3 − 2s4 )OG (x1 x3 ) − (s1 s3 − 2s4 )OG (x1 x4 ) − OG (x31 x22 x3 ) OG (x31 x3 x24 ) = −(s21 s4 − s1 s2 s3 + s23 ) − s1 s3 OG (x1 x3 ) − s2 OG (x21 x2 x4 ) − OG (x31 x22 x3 ) OG (x31 x2 x23 ) = −(s21 s4 − s1 s2 s3 + 2s2 s4 + s23 ) − 2s4 OG (x1 x4 ) − s2 OG (x21 x2 x4 ) − s2 OG (x21 x3 x4 ) − OG (x31 x22 x3 )

4.7 The Ring of Invariants of the Regular Representation of the Klein Group

71

OG (x31 x23 x4 ) = −(s1 s2 s3 − 4s2 s4 ) + (s1 s3 − 2s4 )OG (x1 x3 ) − 2s4 OG (x1 x4 ) + s2 OG (x21 x2 x4 ) + s2 OG (x21 x3 x4 ) + OG (x31 x22 x3 ) OG (x31 x2 x24 ) = −(s1 s2 s3 − 2s2 s4 ) − s1 s3 OG (x1 x3 ) + (s1 s3 − 2s4 )OG (x1 x4 ) + s2 OG (x21 x2 x4 ) + OG (x31 x22 x3 ) This shows that K[V ]G is the K[s1 , s2 , s3 , s4 ]-module generated by 1, OG (x1 x3 ), OG (x1 x4 ), OG (x21 x2 x4 ), OG (x21 x3 x4 ), OG (x31 x22 x3 ). Furthermore, applying Proposition 3.1.4, we see that K[V ]G is in fact the free K[s1 , s2 , s3 , s4 ]module generated by these 6 orbit sums and thus that K[V ]G is CohenMacaulay. Next, we turn to the question of a minimal algebra generating set for this example. The above shows that K[V ]G is generated by the 4 primary invariants s1 , s2 , s3 , s4 and the 6 secondary invariants listed above. Of course, 1 is not required in an algebra generating set. Also, since OG (x21 x2 x4 ) = OG (x1 x2 )OG (x1 x4 ) and OG (x21 x3 x4 ) = OG (x1 x3 )OG (x1 x4 ), we see that OG (x21 x2 x4 ) and OG (x21 x3 x4 ) are decomposable invariants and so not part of a minimal algebra generating set. Similarly, the identity OG (x31 x22 x3 ) = s3 s2 s1 − 4s4 s2 + OG (x1 x4 )OG (x1 x3 )2 + s2 OG (x1 x4 )2 − s22 OG (x1 x4 ) − s3 s1 OG (x1 x3 ) + 2s4 OG (x1 x3 ) shows that OG (x31 x22 x3 ) is decomposable. Therefore, K[V ]G is generated by s1 , s2 , OG (x1 x3 ), OG (x1 x4 ), s3 , s4 . Continuing we may write 4s4 = s1 s3 − OG (x1 x2 )OG (x3 x4 ) − OG (x1 x3 )OG (x2 x4 ) − OG (x1 x4 )OG (x2 x3 ). First suppose that the characteristic of K is different from 2. Then 1/4 ∈ K and so we may use the above identity to write s4 as a polynomial in the other 5 invariants. Therefore, K[V ]G is generated by s1 , s2 , OG (x1 x3 ), OG (x1 x4 ), s3 . This is the best we can do. If we could get by with only 4 generators, then K[V ]G would be a polynomial ring. The form of the Hilbert series shows this is not the case. More directly, recall that the Hilbert series begins H(K[V ]G , λ) = 1 + λ + 4λ2 + 5λ3 + 11λ4 + . . . . Of course, we need the linear generator s1 . We also need three quadratic generators in addition to s21 to span the degree 2 invariants. These 4 invariants will generate only a 4 dimensional subspace of K[V ]3 and thus a fourth generator of degree 3 is required. Now suppose that K has characteristic 2. In this case, 6 algebra generators are required. To see this, consider the point v = (1, 1, 1, 1). The first five generators s1 , s2 , OG (x1 x3 ), OG (x1 x4 ), s3 all vanish at v. However s4 (v) = 1. Hence s4 cannot be expressed in terms of the other 5 generators. Using degree arguments it is easy to see that none of the 5 remaining generators can be expressed in the other 4.

72

4 Examples

4.8 The Ring of Invariants of the Regular Representation of C4 We consider the regular representation of the cyclic group G := C4 of order 4 over a field K. We choose a permutation basis for V and we denote the corresponding permutation basis for V ∗ by {x1 , x2 , x3 , x4 }. So a generator σ ∈ G maps x1 to x2 , x2 to x3 , x3 to x4 , and x4 to x1 . As we saw in §3.7.2, the Hilbert series of the ring of invariants K[V ]G in any characteristic is given by   1 1 1 2 H(K[V ]G , λ) = + + 4 (1 − λ)4 (1 − λ2 )2 (1 − λ4 ) = 1 + λ + 3λ2 + 5λ3 + 10λ4 + 14λ5 + 22λ6 + . . . =

1 + λ2 + λ3 + 2λ4 + λ5 (1 − λ)(1 − λ2 )(1 − λ3 )(1 − λ4 )

By G¨ obel’s theorem, we know that K[V ]G is generated by elements of degrees at most 6. As in the case of the regular representation of the Klein group, we analyze the ring of invariants as a module over K[x1 , x2 , x3 , x4 ]Σ4 = K[s1 , s2 , s3 , s4 ], where s denotes the elementary symmetric functions. Thus if K[V ]G is CohenMacaulay, the generators for K[V ]G as a K[V ]Σ4 -module will have degrees 0, 2, 3, 4, 4, and 5. By the proof of G¨ obel’s Theorem (see Remark 4.6.2), we know that the ring of invariants is generated as a module over K[s1 , s2 , s3 , s4 ] by the orbit sums without gaps. In addition to the elementary symmetric functions, the special orbit sums — just as in the case of the regular representation of the Klein group — are associated to the decreasing exponent sequences. (3, 2, 1, 0), (2, 2, 1, 0), (2, 1, 1, 0), (2, 1, 0, 0). Proceeding just as before to write the Σ4 orbit sums of these and the elementary symmetric functions as G orbit sums, we find that there are 19 orbit sums from which to determine module generators: OΣ4 (x1 ) = s1 = OG (x1 ) OΣ4 (x1 x2 ) = s2 = OG (x1 x2 ) + OG (x1 x3 ) OΣ4 (x1 x2 x3 ) = s3 = OG (x1 x2 x3 ) OΣ4 (x1 x2 x3 x4 ) = s4 = OG (x1 x2 x3 x4 ) OΣ4 (x21 x2 ) = OG (x21 x2 ) + OG (x21 x3 ) OΣ4 (x21 x2 x3 ) = OG (x21 x2 x3 ) + OG (x21 x2 x4 ) + OG (x21 x3 x4 ) OΣ4 (x21 x22 x3 ) = OG (x21 x22 x3 ) + OG (x21 x2 x23 ) + OG (x21 x22 x4 )

4.8 The Ring of Invariants of the Regular Representation of C4

73

OΣ4 (x31 x22 x3 ) = OG (x31 x22 x3 ) + OG (x31 x22 x4 ) + OG (x31 x2 x23 ) + OG (x31 x23 x4 ) + OG (x31 x2 x24 ) + OG (x31 x3 x24 ) In the Cohen-Macaulay case, we will see that 1, OG (x1 x2 ), OG (x21 x2 ), OG (x21 x2 x3 ), OG (x21 x2 x4 ), OG (x21 x22 x3 ) generate K[V ]G as a K[s1 , s2 , s3 , s4 ]-module. Here are the identities needed to verify this: OG (x1 x3 ) = s2 − OG (x1 x2 ) OG (x21 x3 ) = (s1 s2 − s3 ) − s1 OG (x1 x2 ) OG (x21 x3 x4 ) = (s1 s3 − 4s4 ) − OG (x21 x2 x3 ) − OG (x21 x2 x4 ) OG (x21 x2 x23 ) = (−s1 s4 + s2 s3 ) − s3 OG (x1 x2 )) OG (x21 x22 x4 ) = −2s1 s4 + s3 OG (x1 x2 ) − OG (x21 x22 x3 ) OG (x31 x22 x4 ) = (−s21 s4 + 2s2 s4 ) − s4 OG (x1 x2 ) + s2 OG (x21 x2 x4 ) − s1 OG (x21 x22 x3 ) + OG (x31 x22 x3 ) OG (x31 x3 x24 ) = (−s21 s4 − s23 ) − s1 s3 OG (x1 x2 ) − s2 OG (x21 x2 x4 ) − OG (x31 x22 x3 ) OG (x31 x2 x23 ) = (−s21 s4 + 2s2 s4 − s23 ) + s4 OG (x1 x2 ) + s2 OG (x21 x2 x3 ) − OG (x31 x22 x3 ) OG (x31 x23 x4 ) = (s1 s2 s3 − 2s2 s4 ) + (−s1 s3 + s4 )OG (x1 x2 ) − s2 OG (x21 x2 x3 ) + OG (x31 x22 x3 ) OG (x31 x2 x24 ) = (2s2 s4 − s23 ) − s4 OG (x1 x2 ) + s1 OG (x21 x2 x3 ) − OG (x31 x22 x3 ) 2OG (x31 x22 x3 ) = s3 OG (x21 x2 ) − s2 OG (x21 x2 x4 ) + s1 OG (x21 x2 x3 ) Thus if the characteristic of K is not 2, then the 6 module generators listed above suffice. However, in characteristic 2, the last equation yields a K[s1 , s2 , s3 , s4 ] linear relation s3 OG (x21 x2 ) − s2 OG (x21 x2 x4 ) + s1 OG (x21 x2 x3 ) = 0 among the module generators. By considering the Hilbert series we see that in light of this relation, we need an extra module generator which we may take to be OG (x31 x22 x3 ). Thus, in characteristic 2, K[V ]G is not Cohen-Macaulay and requires 7 secondary generators. Now we consider algebra generating sets. The identity s1 s3 − s2 OG (x1 x2 ) + OG (x1 x2 )2 − OG (x21 x2 x4 ) = 4s4 rewrites as

74

4 Examples

OG (x21 x2 x4 ) = s1 s3 − s2 OG (x1 x2 ) + OG (x1 x2 )2 − 4s4 . Hence OG (x21 x2 x4 ) is not required as a generator (in the presence of s4 ). Also, 2OG (x21 x22 x3 ) = −2s1 s4 + s1 OG (x21 x2 x3 ) − s2 s3 − s2 OG (x21 x2 ) + 2s3 OG (x1 x2 ) + OG (x1 x2 )OG (x21 x2 ) which shows that OG (x21 x22 x3 ) is decomposable if the characteristic of the field is not 2. Together with the above relations these two identities show that F[V ]G is generated as an algebra by s1 , s2 , s3 , s4 , OG (x1 x2 ), OG (x21 x2 ), OG (x21 x2 x3 ) and (if the characteristic of K is 2) OG (x21 x22 x3 ), OG (x31 x22 x3 ) . Thus in characteristics different from 2, we see that K[V ]G = K[s1 , s2 , OG (x1 x2 ), s3 , OG (x21 x2 ), OG (x21 x2 x3 ), s4 ]. Note that the largest required generator has degree 4 which is the order of G. In characteristic 2, we have s21 s4 + s23 + OG (x1 x2 )OG (x21 x2 x3 ) = OG (x31 x22 x3 ). Thus OG (x31 x22 x3 ) is not required as an algebra generator even in characteristic 2. Thus, in characteristic 2, K[V ]G is generated by  s1 , s2 , OG (x1 x2 ), s3 , OG (x21 x2 ), s4 , OG (x21 x2 x3 ), OG (x21 x2 x4 ), OG (x21 x22 x3 ) . From the Hilbert series, we know that dimK [V ]G 5 = 14. However, the extra relation in characteristic 2 given by s1 OG (x21 x2 x3 ) − s2 s3 − s2 OG (x21 x2 ) + OG (x1 x2 )OG (x21 x2 ) = 0 (which follows from the above identities) shows that the 8 generators other than OG (x21 x22 x3 ) only generate a 13 dimensional subspace of K[V ]5 . Thus an algebra generator of degree 5 is required in characteristic 2. Remark 4.8.1. The example of the regular representation of C4 over the field F2 of order 2 is important historically. Pierre Samuel asked the question whether every unique factorization domain was Cohen-Macaulay. His student Marie-Jos´e Bertin [9] answered this question in the negative using the example we have just considered. She showed that this ring of invariants is not Cohen-Macaulay and since it is the ring of invariants of a p-group, it is a unique factorization domain. For a good description of this, see [43].

4.10 The Three Dimensional Modular Representationof Cp

75

4.9 A 2 Dimensional Representation of C3 , p = 2 We consider a representation V of G = C3 with generator σ on V of dimension 2 over F2 . We suppose that the action of σ on V ∗ with basis {x, y} given   01 σ= , 11 Then σ acting on F2 [V ] = F2 [x, y] sends x to y and sends y to x + y. It is straightforward to calculate the ring of invariants for G. First we observe that the Dickson invariants (see §3.3) r = x2 + xy + y 2 , s = x2 y + xy 2 form, as always, a homogeneous system for F2 [V ]G . Second we observe that G has index 2 in GL2 (F2 ), and that t = x3 + x2 y + y 3 is invariant. It isn’t hard to see using Galois theory that F2 [x, y]G = F2 [x2 + xy + y 2 , x2 y + xy 2 , x3 + x2 y + y 3 ], and, therefore, this ring is a hypersurface. As part of this calculation, we note t2 = r3 + s2 + rs. Therefore, we obtain a resolution over the ring A = F2 [a, b, c] with deg(a) = 2, deg(b) = deg(c) = 3, and ρ(a) = x2 + xy + y 2 , ρ(b) = x2 y + xy 2 , ρ(c) = x3 + x2 y + y 3 . We obtain 0 → A(c2 + a3 + b2 + bc) → A → F2 [x, y]G → 0. It follows that H(F2 [V ]G , t) =

1 + t2 + t4 . (1 − t3 )2

4.10 The Three Dimensional Modular Representation of Cp Suppose p > 2 and let F be a field of characteristic p. Consider the action of Cp = σ on a three dimensional F vector space V determined by the matrix ⎛ ⎞ 100 σ = ⎝1 1 0⎠ . 011 Note that this matrix has order p and thus does indeed afford a three dimensional representation of G = Cp . This representation turns out to be the unique indecomposable representation of Cp of this dimension. We will discuss this question and related matters later in §7.1. We will compute the ring of invariants of this representation and see that it is a hypersurface. We will give two different computations of this ring of invariants. The first takes advantage of prior knowledge of the Hilbert series

76

4 Examples

which reduces the amount of work considerably. The second is longer but does not rely on any prior knowledge. We let {x, y, z} denote the basis of V ∗ dual to the standard basis of V . We begin by observing that some invariants are easily constructed: x is invariant, as are NG (y) and NG (z). Furthermore, it is not hard to see that d = y 2 − 2xz − xy is also invariant. You may well wonder where d came from. There are two considerations, one coming from the form of the Hilbert series, see immediately below, and the other from a consideration of what we refer to as “integral invariants”, see §7.5. In the meantime, we note simply that while the polynomial expressions for NG (y) and NG (z) depend upon p, the expressions for x and d do not: the “same” polynomial x, or d, is invariant for all primes. Theorem 4.10.1. F[V ]G = F[x, NCp (y), NCp (z), d] The next two sections are devoted to the two proofs of this theorem. 4.10.1 Prior Knowledge of the Hilbert Series Our first approach to this example uses the work of Almkvist and Fossum [3]. They have shown that there exists a three dimensional representation of Cp in characteristic 0 which has the same Hilbert series as the current ring of invariants F[x, y, z]Cp . Let ξ be a complex (primitive) p-th root of unity and consider the matrix ⎛ −2 ⎞ ξ 0 0 τ = ⎝ 0 1 0⎠. 0 0 ξ2 For p > 2, the matrix τ has order p and the Hilbert series of the associated representation of Cp can be shown to be the same as the Hilbert series of the characteristic p representation. Almkvist and Fossum’s work has been extended and explained by Hughes and Kemper in references [54] and [55]. Lemma 4.10.2. We have p−1

H(F[V ] , λ) = G

λ2i 1 + λp = . (1 − λ)(1 − λp )2 (1 − λ)(1 − λ2 )(1 − λp ) i=0

  The first form of the Hilbert series suggests that there is a homogeneous system of parameters for F[V ]G consisting of one invariant of degree 1 and two of degree p, and that there is a fourth invariant of degree 2, whose powers give module generators for the ring of invariants over the subalgebra generated by the homogeneous system. The second form of the Hilbert series suggests that there is a homogeneous system of parameters for F[V ]G consisting of one invariant of degree 1, one

4.10 The Three Dimensional Modular Representationof Cp

77

invariant of degree 2 and one invariant of degree p, and that there is a fourth invariant of degree p which generates the ring of invariants as a module over the subalgebra generated by the homogeneous system. Both of these suggestions turn out to be correct, but the reader should be warned that the form of the Hilbert series cannot always be “realized”. It is easy to find three invariants: first of all x is invariant, and we may form NG (y) = y p − xp−1 y and NG (z) = z p − xz p−1 − . . . . We note that NG (y) is F-linear in y, but that NG (z) is not. Consideration of this phenomenon eventually leads to the following. Lemma 4.10.3. 1. Let x ∈ V ∗ and suppose G is a p-group acting on V ∗ with the property that (G − 1)x = {(σ − 1)(x) | σ ∈ G} is a vector space. Then NG (x) is a p-polynomial, that is, is the only powers t of x which occur are of the form xp and hence the polynomial is Fp -linear in x. 2. The coefficients of the various powers of x in the polynomial NG (x) are the Dickson invariants in the elements of the vector space (G − 1)(x). 3. NG (x) is F-linear in x if and only if (G − 1)(x) is a vector space.   By Proposition 4.0.3, the sequence x, N(y), N(z) is a homogeneous system of parameters for F[V ]Cp . Therefore, F[V ]G is finitely generated as a module over the subalgebra A = F[x, N(y), N(z)]. As noted above, the form of the Hilbert series suggests the existence of an invariant of degree 2. We note that F[V ]2 has dimension 6, and it is fairly easy to compute the action of σ on F[V ]2 and determine that d = y 2 − 2xz − xy and x2 span the homogeneous invariants of degree 2. We also note that the proof of Proposition 4.0.3 shows that {x, d, N(z)} is also a homogeneous system of parameters for F[V ]Cp . What relation holds between the generators above? We observe that dp and N(y)2 both have terms of the form y 2p and this allows the one to be written as a polynomial expression in x, N(z) and the other. Further, no lower power of di can be written as a polynomial in x, N(y) and N(z) given the term y 2i in di . Thus i F[x, d, N(y), N(z)] = ⊕p−1 i=0 F[x, N(y), N(z)]d .

/ F[x, d, N(z). Hence Similarly, N(y)2 ∈ F[x, d, N(z) but N(y) ∈ F[x, d, N(y), N(z)] = F[x, d, N(z)] ⊕ F[x, d, N(z)]N(y) Using either of these decompositions we can find H(F[x, d, N(y), N(z)], λ). For example, from the second decomposition, we see

78

4 Examples

H(F[x, d, N(y), N(z)], λ) = H(F[x, d, N(z)], λ)(1 + λp ) 1 + λp = . (1 − λ)(1 − λ2 )(1 − λp ) Given the computation of Almkvist and Fossum above which computes the Hilbert series of F[V ]G , we see H(F[x, d, N(y), N(z)], λ) = H(F[V ]G , λ). Since F[x, d, N(y), N(z)] ⊆ F[V ]G , these two rings must be equal. We contrast the preceding with the computation below in which none of the individual steps are hard, but there are many steps to take. 4.10.2 Without the Use of the Hilbert Series Without the information encoded in the Hilbert series we must work harder to compute the ring of invariants. As above, it is easy to construct invariants, but it is much more difficult to know when to stop. That is, it is hard to determine which sets of invariants are generating sets in the absence of other information. The next theorem tells us that the only invariants in x, y and z which have degree less than p in z must be polynomials in x, d and NG (y) = N(y). j Theorem 4.10.4. If f ∈ F[V3 ]Cp and f = i=0 fi z i where fi ∈ F[x, y, d] and j < p, then f ∈ F[x, d, N(y)]. Proof. The proof of the theorem is by induction on j. For j = 0, f = f0 ∈ F[x, y, d]. Write f=

n 

a i di

i=0

where ai ∈ F[x, y] and apply delta: 0 = Δ(f ) =

n 

Δ(ai )di .

i=0

Since {x, y, d} is algebraically independent, Δ(ai ) = 0. Thus ai ∈ F[V2 ]Cp = F[x, N(y)]. For j > 0, we apply delta to f to get 0 = Δ(f ) = Δ(fj )z j + σ(fj )Δ(z j ) + . . . Since degz (Δ(z j )) < j, we have Δ(fj ) = 0. Rewriting Δ(f ) gives

4.10 The Three Dimensional Modular Representationof Cp

79

0 = Δ(fj )σ(z j ) + fj Δ(z j ) + Δ(fj−1 )z j−1 + . . . or 0 = jfj yz j−1 + Δ(fj−1 )z j−1 + terms with smaller z-degree. Thus fj ∈ Δ(F[x, y, d]) ⊂ (x)F[x, y, d]. Write fj = xh with h ∈ F[x, y, d]. Let d = d − zx ∈ F[x, y] so that zx = d − d . Then fj z j = h(xz)z j−1 = hdz j−1 − hd z j−1 and f = (hd − hd + fj−1 )z j−1 +

j−2 

fi z i .

i=0

Thus, by induction, f ∈ F[x, d, N(y)].

 

Since N(z) is monic when considered as a polynomial in the variable z, we may divide any polynomial f by N(z) to get f = qN(z) + r where q and r are unique with degz r < p. Lemma 4.10.5. If f ∈ RG and f = qN(z) + r with degz r < p, then q, r ∈ RG . Proof. We have f = σ ·f = (σ ·q)(σ ·N(z))+(σ ·r) = (σ ·q)N(z)+(σ ·r). Since σ · z = z + y, σ · y = y + x and σ · x = x, it follows that degz (σ · h) = degz (h) for all h ∈ R. In particular, degz (σ · r) = degz (r) < p. Thus by the uniqueness of remainders and quotients, we must have σ · r = r and σ · q = q. Hence   σ i · r = r and σ i · q = q and thus q, r ∈ RG . Consider f ∈ F[x, y, z]G and write f = q · N(z) + r where degz (r) < p. We may also write p−1  r= ri z i i=0

where ri ∈ F[x, y]. Before proceeding to the proof of Theorem 4.10.1, we have the following lemma. Lemma 4.10.6. xi divides ri . Proof. r = σ(r) =

p−1 

σ(ri )σ(z i )

i=0

=

p−1  i=0

σ(ri )

i    i j=0

j

y i−j z j

80

4 Examples

Fix j with 0 ≤ j ≤ p−1. Equating the coefficients of z j in the above equations we get p−1    i i−j rj = y σ(ri ). j i=j Equivalently, we have p−1    i i−j (j + 1)yσ(rj+1 ) = −(σ − 1)(rj ) + y σ(ri ) j i=j+2

Applying σ −1 to this equation we obtain: (j + 1)σ

−1

(y)rj+1 = −(σ

−1

p−1    i −1 i−j σ (y )(ri ) − 1)(rj ) + j

(4.10.1)

i=j+2

Note that σ −1 (y) = y − x and σ −1 (x) = x. Thus if for some k we have xk divides rj , then we also have xk+1 divides (σ −1 − 1)(rj ). Therefore, if xk+1 divides ri for all i = j+2, . . . , p−1 and also xk divides rj , then Equation 4.10.1 implies that xk+1 divides rj+1 . We proceed by induction. We will prove that xt divides rt and that xt+1 divides ri for all i = t + 2, . . . , p − 1 by induction. For the initial step, t = 0, we consider Equation 4.10.1 with j = p − 2. From that equation, we see that x divides rp−1 . Then the above remark with k = 0 and j = p − 3 shows that x divides rp−2 . Continuing to apply this remark we obtain that x divides ri for all i = 1, . . . , p − 1 which completes the initial step of the induction. For the general step, we assume that xt−1 divides rt−1 and that xt divides ri for all i = t + 1, . . . , p − 1. But then using the above remark again with k = t − 1 and j = t − 1, we obtain that xt divides rt . Now applying the remark with k = t and j = p − 2 shows that xt + 1 divides rp−2 . Next applying the remark with k = t and j = p − 3 shows that xt + 1 divides rp−3 . Continuing in this fashion we obtain that xt+1 divides ri for all i = t + 1, . . . , p − 1 which completes the proof of the lemma.   We now give the proof of Theorem 4.10.1. Proof. As above, we take f ∈ F[x, y, z]G and write f = qN(z)+r. We consider elements of F[x, y, z] as polynomials in z. In particular, d = xz + (

p−1 2 p+1 y + xy). 2 2

By the lemma, we may divide d into r to obtain r = d·g+h where degz (h) = 0. Then r = σ(r) = d · σ(g) + σ(h) with degz (σ(h)) = 0. Thus by the uniqueness of the remainder, we see that g = σ(g) and h = σ(h), i.e., g, h ∈ F[x, y, z]G .

4.10 The Three Dimensional Modular Representationof Cp

81

In summary, we have shown that if f ∈ F[x, y, z]G , then we have f = q · N(z) + r = q · N(z) + d · g + h where h ∈ F[x, y]G = F[x, N(y)]. By induction on degree we may show that g, q ∈ F[x, d, N(y), N(z)] and thus h ∈ F[x, d, N(y), N(z)].  

5 Monomial Orderings and SAGBI Bases

Let S denote the polynomial ring K[x1 , . . . , xn ]. The set of monomials in S is M := {xa1 1 xa2 2 · . . . · xann | a1 , a2 , . . . , an ∈ N}. A term in S is an element of the form cm where c ∈ K and m ∈ M. Definition 5.0.1. A monomial order is a total order on M such that 1. m > 1 for all m ∈ M \ {1}, and 2. if m1 > m2 , then mm1 > mm2 for all monomials m, m1 , m2 . The following lemma shows that every monomial ordering is a wellordering, i.e., every non-empty set of monomials contains a least element. This is the key fact we need in order to use monomial orders as the basis of induction proofs. Lemma 5.0.2. A monomial ordering is a well-ordering. Proof. Consider a non-empty set of monomials A ⊆ M and let I be the ideal of S generated by A. Then the monomials in I are precisely those monomials in S which are divisible by some monomial of A. Since S is a Noetherian ring, the ideal I is minimally generated by some finite subset {m1 , . . . , mr } of A. Without loss of generality we assume that m1 is the smallest monomial in the finite set {m1 , . . . , mr }. Since {m1 , . . . , mr } generates I, the set of monomials in I is also characterized as those monomials of S which are divisible by some mi with 1 ≤ i ≤ r. Take any monomial m in A. Then m lies in I and so there exists i such that mi divides m. Write m = m mi . Then 1 ≤ m . This together with m1 ≤ mi implies m1 ≤ m m1 ≤ m mi = m. Therefore, m1 is less than or equal to every monomial of A.   Now let f be any non-zero element of S and write f (uniquely) as a linear combination of distinct monomials: f = c1 m1 + c2 m2 + . . . + cr mr where ci ∈ K and mi ∈ M for all i = 1, 2, . . . , r. Without loss of generality, assume m1 > mi for all i = 2, 3, . . . , r. We say that m1 is the lead monomial of f and write LM(f ) = m1 . Similarly, we say that c1 m1 is the lead term of f and write H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 5, © Springer-Verlag Berlin Heidelberg 2011

83

84

5 Monomial Orderings and SAGBI Bases

LT(f ) = c1 m1 , that c1 is the lead coefficient of f and write LC(f ) = c1 . We make the conventions that LM(0) = 0, LT(0) = 0 and LC(0) = 0 although we do not consider 0 a term nor a monomial. We now give some examples of monomial orderings. In these examples we a a a will use two different monomials m = xa1 1 xa2 2 ·. . .·xann and m = x1 1 x2 2 ·. . .·xnn . Example 5.0.3 (Lexicographic Ordering). Define i ≥ 1 by a1 = a1 , a2 = a2 , . . . , ai−1 = ai−1 but ai = ai . Then in the lexicographic ordering, m mp − (n1 + n2 + · · · + nm ), then some di must exceed p − ni and therefore, F[V ]d is free. This proves following result. Proposition 7.3.3. Suppose V is a reduced Cp -module with m = dim V Cp .  Let d be a positive integer with d > m(p − 2). Then F[V ]d ∼ = F[V ]d ⊕ k Vp for some k.

7.4 The First Fundamental Theorem for V2 This section is devoted to finding generators for the vector invariants of V2 , i.e., for F[m V2 ]Cp for all m ≥ 1. In 1990, Richman [92] conjectured a set of generators for F[m V2 ]Cp . Campbell and Hughes [19] proved Richman’s conjectured generating set is correct in 1997. Their proof itself is somewhat difficult and relies upon a deep result of Wilson concerning the rank of 0-1 matrices in characteristic p. Here we give a new proof of the first fundamental theorem due to Campbell, Shank and Wehlau [21]. This new proof enjoys a number of advantages. The new proof yields a description of the Cp -module structure of F[m V2 ]. This description is quite explicit and easily computable. The new proof also provides a SAGBI basis for the ring of invariants. Thirdly, the new proof is simpler and in particular avoids the use of Wilson’s theorem. We will prove the following.

116

7 The Cyclic Group Cp

Theorem 7.4.1. Let G = Cp act on V = m V2 . Let {yi , xi } denote a basis for the ith copy of V2∗ in V ∗ where σ(yi ) = yi + xi and σ(xi ) = xi . The ring of invariants F[m V2 ]Cp is generated by the following invariants: 1. xi for i = 1, 2, . . . , m. 2. N(yi ) = yip − xip−1 yi for i = 1, 2, . . . , m. 3. uij = xi yj − xj yi for 1 ≤ i < j ≤ m. am 4. TrCp (y1a1 y2a2 . . . ym ) where 0 ≤ ai < p for i = 1, 2, . . . , m. Moreover, this set of generators is a SAGBI basis for the ring of invariants with respect to the graded reverse lexicographic order determined by y1 > x1 > y2 > · · · > xm . A theorem such as this giving explicit algebra generators for all vector invariants of a fixed representation is called a first fundamental theorem. A second fundamental theorem would give a generating set for the algebraic relations among such generators. Remark 7.4.2. It was shown by Shank and Wehlau [99] that the invariants am of the form TrCp (y1a1 y2a2 . . . ym ) with a1 + a2 + · · · + am ≤ 2(p − 1) are not required in a minimal generating set, nor in a SAGBI basis. However, none of the other invariants listed in Theorem 7.4.1 can be omitted. Proof. We now begin the proof of Theorem 7.4.1. Let A denote the ring generated by the invariants given in the statement, A := F[xi , N(yi ), uij , Tr(y A ) | 1 ≤ i < j ≤ m, 0 ≤ ai < p] . C

We need to show that F[m V2 ]d p ⊆ Ad for all d ≥ 0. We proceed by induction on d. The case d = 0 is clear since F[m V2 ]0 = F. Now consider the genC eral case d > 0. Take f ∈ F[m V2 ]d p and write f = f  + f  where  C p f = m i=1 fi N(yi ) with fi ∈ F[m V2 ]d−p . By induction, each of the fi ∈ A  and thus f ∈ A. Hence we may assume from now on that f = f  . Suppose C f = f  ∈ F[m V2 ](dp1 ,d2 ,...,dm ) where d1 + d2 + · · · + dm = d. If there exists i with di ≥ p then since degyi (f ) < p we see that xi divides f . Hence f = xi f  p where f  ∈ F[m V2 ]d−1 . Again, by induction, this implies that f  and hence f lies in A. Therefore, we may assume that di < p for all i = 1, 2, . . . , m. We now recall the polarization and restitution operators defined in §1.9. We will take f ∈ F[m V2 ]d and consider its full polarization P(f ) = f(1,1,...,1) . We will also exploit the restitution map

C

R = R(d1 ,d2 ,...,dm ) : F[d V2 ](1,1,...,1) → F[m V2 ](d1 ,d2 ,...,dm ) defined by R(f ) = f (v1 , . . . , v1 , v2 , . . . , v2 , . . . , vm , . . . , vm )    

  d1

d2

dm

7.4 The First Fundamental Theorem for V2

117

where d = d1 + d2 + · · · + dm . We have R(P(f )) = λf where λ = d1 !d2 ! · · · dm !. Since each di < p we see Cp that f = R(F ) where F = P(1,1,...,1) (λ−1 f )) ∈ F[d V2 ](1,1,...,1) . We will show that f ∈ Ad by considering the lead monomials of f and F . We are using {xj , yj } to denote the usual co-ordinate functions for the j th copy of V2 in m V2 . We have F[d V2 ] = F[d1 V2 ⊕ d2 V2 ⊕ · · · ⊕ dm V2 ]. We let {xj1 , yj1 , xj2 , yj2 , . . . , xjdj , yjdj } denote the co-ordinate functions on dj V2 . We will use the graded reverse lexicographic order on F[m V2 ] with y1 > x1 > y2 > x2 > · · · > ym > xm . Similarly, we use the graded reverse lexicographic order on d V2 determined by y11 > x11 > y12 > x12 > · · · > x1d1 > y21 > · · · > xddm Thus xjk < yjk , x(j+1)k < xjk , y(j+1)k < yjk , xj(k+1) < xjk and yj(k+1) < yjk . With this term order we have the following lemma. Lemma 7.4.3. Let d1 , d2 , . . . , dm be non-negative integers with 0 ≤ di < p Cp for i = 1, 2, . . . , m. Suppose F ∈ F[d V2 ](1,1,...,1) is a multi-linear invariant of degree d = d1 + d2 + · · · + dm . Then LM(F ) ∈ {LM(P(f ) | f ∈ A(d1 ,d2 ,...,dm ) }. We defer the proof of Lemma 7.4.3 until §7.4.2. Instead, we show why Lemma 7.4.3 suffices to prove the invariants listed in Theorem 7.4.1 do form a generating set for the ring of invariants. To see this, assume by way of contradiction, that f is a homogeneous element of F[m V2 ]Cp \ A. By the above, we have that f = f  is homogeneous of multi-degree (d1 , d2 , . . . , dm ) where di < p for all i = 1, 2, . . . , m. Suppose f is such that LM(P(f )) is minimal among those f = f  ∈ F[m V2 ]Cp \ A. Write f = RP(λ−1 f ) where λ = d1 !d2 ! · · · dm ! ∈ F. By Lemma 7.4.3, there exists h ∈ A(d1 ,d2 ,...,dm ) with LT(P(h)) = LT(P(λ−1 f )). Define F  := P(λ−1 f ) − P(h). Therefore, we have LM(F  ) = LM(P(λ−1 f −h)) < LM (P(λ−1 f )). By the choice of f , this implies that λ−1 f − h ∈ A. Thus f = λ(λ−1 f − h) + λh ∈ A. This contradiction shows that the invariants listed in the statement of Theorem 7.4.1 do indeed form a generating set for the ring of invariants.   We will prove that these invariants are not just a generating set but indeed a SAGBI basis at the end of §7.4.2. 7.4.1 Dyck Paths and Multi-Linear Invariants Before we can prove Lemma 7.4.3, we need to describe the multi-linear Cp invariants. This amounts to giving a description of the decomposition of the

118

7 The Cyclic Group Cp

Cp -module ⊗d V2 into indecomposable Cp -modules. Each such summand contains a fixed line and these are the multi-linear invariants we seek. Define non-negative integers μdp (k) by the direct sum decomposition of the Cp -module ⊗d V2 over Fp : d 

V2 ∼ =

p 

μdp (k) Vk .

k=1

We have the following lemma. Lemma 7.4.4. Let p ≥ 3. Then μ0p (k) = δk1 and μ1p (k) = δk2 , and

⎧ d μp (2), ⎪ ⎪ ⎪ ⎨μd (k − 1) + μd (k + 1), p p (k) = μd+1 p d ⎪ μ (p − 2), ⎪ p ⎪ ⎩ d μp (p − 1) + 2μdp (p),

if if if if

k = 1; 2 ≤ k ≤ p − 2; k = p − 1; k = p;

for d ≥ 1. Proof. The initial conditions are clear. The recursive conditions follow immediately from the three equations from Lemma 7.1.14: V1 ⊗ V2 = V2 Vk ⊗ V2 = Vk−1 ⊕ Vk+1 for all 2 ≤ k ≤ p − 1 V p ⊗ V2 = 2 V p .   We now introduce certain combinatorial objects that we will use to describe the decomposition of ⊗d V2 into indecomposable summands. At the end of this section, we provide an example illustrating the correspondence between invariants and these combinatorial objects. A Dyck path of length d is a lattice path in the first quadrant of the xyplane from (0,0) to (d, 0) comprised of a sequence of steps each of the form (1,1) or (1,-1). Steps of the form (1,1) are called rises and denoted by R. Steps of the form (1, −1) are called falls and denoted by F . A Dyck path is encoded by a Dyck word which is a sequence of R’s and F ’s with an equal number of each. The condition that the path remain in the first quadrant, i.e., on or above the x-axis corresponds to the restriction on the word that among the first t symbols there are never more F ’s than R’s for all 1 ≤ t ≤ d. If Γ = γ1 γ2 . . . γd is a Dyck word, then for each t with 1 ≤ t ≤ d, the initial portion of the word, γ1 γ2 . . . γt , is a partial Dyck word and the corresponding t steps are called a partial Dyck path.

7.4 The First Fundamental Theorem for V2

119

The path height or height of a (partial) Dyck path is q if the path touches but does not cross the line y = q. We will say that a partial Dyck path from (0,0) to (t, h) has finishing height h. An initially Dyck path of escape height q and length d is a partial Dyck path from (0,0) to (t, q) for some t ≤ d followed by an entirely arbitrary sequence of d − t steps each step being a rise or a fall. Thus an initially Dyck path may wander after touching the line y = q and in particular is not confined to the first quadrant. We denote the set of initially Dyck paths of escape height q and length d by IDPdq We let PDPd≤q denote the set of all partial Dyck paths of length d and height at most q. We write PDPd≤q (h) to denote the set of partial Dyck paths of length d, height at most q and finishing height h. We will abuse notation and terminology occasionally by identifying a Dyck path with its corresponding word and vice versa. We will denote the set of all finite words (without any restrictions) that can be formed using the alphabet {R,F} by {R,F}∗ . Let νqd (h) := |PDPd≤q (h)| for 1 ≤ h ≤ q. We also define ν¯qd := |IDPdq |. With this notation we have the following lemma. Lemma 7.4.5. Let q ≥ 2. Then νq0 (h) = δh0 and νq1 (h) = δh1 , ν¯q0 (h) = 0 and ν¯q1 (h) = 0 , and

⎧ d ⎪ ⎨νq (1), νqd+1 (h) = νqd (h − 1) + νqd (h + 1), ⎪ ⎩ d νq (q − 1),

if h = 0; if 1 ≤ h ≤ q − 1; if h = q;

and d ν¯qd+1 = νq−1 (q − 1) + 2¯ νqd

for all d ≥ 1. Proof. All of these equations are easily seen to hold except perhaps the final one. Its left-hand term ν¯qd+1 = |IDPd+1 | is the number of initial Dyck paths of q length d + 1 and escape height q. We divide such paths into two classes: those which first achieve height q on their final step and those which achieve height q sometime during the first d steps. Paths in the first class are partial Dyck paths of length d, height at most q − 1 and finishing height q − 1 followed by d a rise on the (d + 1)st step. There are νq−1 (q − 1) = |PDPd≤q−1 (q − 1)| such paths. The second class consists of initial Dyck paths of escape height q and length d followed by a final step which may be either a rise or a fall. Clearly, νqd paths of this kind.   there are 2|IDPdq | = 2¯

120

7 The Cyclic Group Cp

Corollary 7.4.6. For all d ∈ N, all primes p ≥ 2 and all k = 1, 2, . . . , p − 1 we have d d (k − 1) and μdp (p) = ν¯p−1 . μdp (k) = νp−2 Proof. Comparing the recursive expressions and initial conditions for μdp (k) d d and νp−2 (k − 1) and for μdp (p) and ν¯p−1 given in the previous two lemmas makes the result clear for p ≥ 5. For p = 2, it is easy to see that μd2 (1) = ν0d (0) = δd0 for d ≥ 0 and d μ2 (2) = 2d−1 = ν¯1d for d ≥ 1.

1, if k + d is odd; d d For p = 3 and k = 1, 2, we have μ3 (k) = ν1 (k −1) = 0, if k + d is even. Hence μd3 (3) =  2 3−1  for d ≥ 0. From the recursive relation ν¯2d+1 = ν1d (1) + d   2¯ ν2d we see that ν¯2d =  2 3−1  = μd3 (3). d

Given an arbitrary word Γ = γ1 γ2 · · · γt ∈ {R,F}∗ we define a multi-linear monomial Λ(Γ ) = z1 z2 · · · zt ∈ F[t V2 ] by taking zi = xi if γi = R and zi = yi if γi = F. We will show that for k = 1, 2, . . . , p − 1, each element of Λ(PDPd≤p−2 (k)) is the lead term of an invariant which spans the fixed line of a summand isomorphic to Vk of ⊗d V2 . Further, we will show that each element of Λ(IDPdp−1 ) is the lead term of an invariant in the image of the transfer. Consider a partial Dyck path of length d, height at most p−1 and finishing height k. Let Γ = γ1 γ2 · · · γd denote the corresponding partial Dyck word. We wish to match each of the falls γj with a rise γπ(j) . We do this recursively as follows. Choose the smallest value of j such that γj is a fall which we have not yet matched with a rise. Let i be maximal such that i < j, γi = R and i = π() for all  < j. Then we declare that π(j) = i. Let I1 := {j | γj = F}, I2 := π(I1 ) and I3 := {1, 2, . . . , d} \ (I1 ∪ I2 ). Note that |I3 | = k and γi = R   for all i ∈ I3 . We define θ(Γ ) = j∈I1 uπ(j),j i∈I3 xi . We also define   θ  (Γ ) = j∈I1 uπ(j),j i∈I3 yi . This construction associates to each partial Dyck path of length d, height at most p − 1 and finishing height k a multi-linear invariant θ(Γ ) ∈ F[d V2 ]Cp . Also, ⎞ ⎛   LM(uπ(j),i )⎠ xi LM(θ(Γ )) = ⎝ ⎛ =⎝

j∈I1

 j∈I1

⎞ xπ(j) yj ⎠

i∈I3



xi

i∈I3

= Λ(Γ ). Next, we suppose Γ = γ1 γ2 · · · γd is a word corresponding to a initially Dyck path of escape height p − 1 and length d. Then there exists t with 1 ≤

7.4 The First Fundamental Theorem for V2

121

t ≤ d such that Γ  := γ1 γ2 · · · γt corresponds to a partial Dyck path of finishing  height p − 1. Let t be minimal d with this property and let Γ := γ1 γ2 · · · γt .   We define θ(Γ ) := Tr(θ (Γ ) i=t+1 zi ) where zi = xi if γi = R and zi = yi if γi = F. Thus we have associated a multi-linear invariant θ(Γ ) ∈ TrCp (F[d V2 ]) to each initially Dyck path of escape height p − 1 and length d. We have ⎞ ⎞ ⎛ ⎛     TrCp (θ (Γ  )) = ⎝ uπ(j),j ⎠ TrCp ( yk ) = ⎝ uπ(j),j ⎠ xk j∈I1

since |I3 | = p − 1. Thus ⎛ LM(TrCp (θ  (Γ  ))) = ⎝



k∈I3

⎞ LM(uπ(j),j )⎠

j∈I1

j∈I1



⎛ xk = ⎝

k∈I3

k∈I3



⎞ xπ(j) yj ⎠

j∈I1



xk .

k∈I3

Therefore, ⎛ LM(TrCp (θ(Γ ))) = ⎝



⎞



xπ(j) yj ⎠

j∈I1

xk

k∈I3

m 

zi = Λ(Γ )

i=t+1

where zi = xi if γi = R and zi = yi if γi = F. We have Proposition 7.4.7. Let γ ∈ PDPd≤p−2  IDPdp−1 . Then 1. LM(θ(γ)) = Λ(γ). Cp ∼ 2. If γ ∈ PDPd≤p−2 (h), then the invariant θ(γ) lies in F[d V2 ](1,1,...,1) = d Cp (⊗ V2 ) and has length h + 1 for all 0 ≤ h ≤ p − 2. Cp ∼ 3. If γ ∈ IDPdp−1 , then the invariant θ(γ) lies in F[d V2 ](1,1,...,1) = (⊗d V2 )Cp and has length p.   Cp ∼ 4. The set θ PDPd≤p−2  IDPdp−1 is a basis for F[d V2 ](1,1,...,1) = (⊗d V2 )Cp . Proof. We have already seen that LM(θ(γ)) = Λ(γ) and that Λ(γ) ∈ Cp . If γ ∈ PDPd≤p−2 (h), then F[d V2 ](1,1,...,1) Δh−1 (θ (Γ )) = Δh−1 =

 j∈I1

 j∈I1

uπ(j),j



     yi = uπ(j),j Δh−1 yi

i∈I3

j∈I1

  uπ(j),j (h − 1)! xi = (h − 1)! θ(Γ ).

i∈I3

i∈I3

Therefore, (θ(Γ )) ≥ h. Since the lead terms of the invariants in the image of θ are distinct, we see that the image of θ is a linearly independent set. This shows that the

122

7 The Cyclic Group Cp

d d image of θ is a basis for (⊗d V2 )Cp since | p−2 h=0 PDP≤p−2 (h)  IDPp−1 | = p−1 d p−2 d d d d Cp ¯p−1 = h=1 μp (h) + μp (p) = dimF (⊗ V2 ) . Furthermore, h=0 νp−2 (h) + ν this equality shows that θ(IDPdp−1 ) is a basis for TrCp (⊗d V2 ) and that each element of θ(PDPd≤p−2 (h)) spans the fixed line of one summand of ⊗d V2 isomorphic to Vh .  

7.4.2 Proof of Lemma 7.4.3 In this section we combine some of the above results in order to prove Lemma 7.4.3. After giving this proof we also prove the final assertion of Theorem 7.4.1. We maintain the notation of the previous two sections. We will use the following lemma which is easy to prove. Lemma 7.4.8. Suppose γ1 and γ2 are two monomials in F[m V2 ](d1 ,d2 ,...,dm ) with γ1 > γ2 . Then LM(P(γ1 )) > LM(P(γ2 )). a

a11 a12 mdm For ease of notation, we will write TrCp (ya ) = TrCp (y11 y12 · · · ymd ) m where a = (a11 , a12 , . . . , amdm ).

Proof (of Lemma 7.4.3). Cp Suppose F ∈ F[d V2 ](1,1,...,1) is a multi-linear invariant of degree d. By Proposition 7.4.7, we may write    xij LM(uij,k ) LM(TrCp (ya )) LM(F ) = (i,j)∈B1

a∈B3

(i,j,k,)∈B2

for some index sets B1 , B2 and B3 . Define    R(xij ) R(uij,k ) R(TrCp (ya )) . f := (i,j)∈B1

a∈B3

(i,j,k,)∈B2

Note that R(xij ) = xi , R(uij,k ) = ui,k (or 0 if i = k). Also, since R is Cp -equivariant, p−1   amdm amdm Cp a11 k a11 R(Tr (y11 · · · ymdm )) = R σ (y11 · · · ymdm ) k=0

 p−1

=

k=0 Cp

= Tr

a

a11 mdm σ k R(y11 · · · ymd ) m a

a11 mdm (R(y11 · · · ymd ) m

αm ) = TrCp (y1α1 y2α2 · · · ym

 i where αi = dj=1 aij ≤ di < p for 1 ≤ i ≤ m. Thus f ∈ A. Let Γ1 := LM(F ). By Lemma 7.4.8, Γ1 = LM(P(γ1 )) = LM(P(f )) where

7.4 The First Fundamental Theorem for V2

γ1 = R(Γ1 ) =

 (i,j)∈B1

xi

 (i,j,k,)∈B2

ui,k



123

TrCp (R(ya )) = LM(f ) .

a∈B3

Hence f ∈ A with LM(F ) = LM(P(f )) as required.

 

Proof (of the second assertion of Theorem 7.4.1). We need to prove that the invariants listed in Theorem 7.4.1 are in fact a SAGBI basis for the ring of invariants. Let f ∈ F[m V2 ]Cp . Dividing f by each N(yi ) we may reduce to the case where f = f  . Furthermore, if f is divisible by any xi , we may replace f C by f /xi . Thus we may reduce to the case where f = f  ∈ F[m V2 ](dp1 ,d2 ,...,dm ) with 0 ≤ di < p for all i = 1, 2, . . . , m. Next, we note that the above proof demonstrates something stronger than the statement given in Lemma 7.4.3. The proof shows that for evC ery f ∈ F[m V2 ](dp1 ,d2 ,...,dm ) , the lead monomial LM(P(f )) lies in the algebra generated by {LM(P(xi ) | 1 ≤ i ≤ m} ∪ {LM(P(uij )) | 1 ≤ i < j ≤ am m} ∪ {LM(P(TrCp (y1a1 y2a2 · · · ym ))) | 0 ≤ ai < p for i = 1, 2, . . . , m}. From this it follows immediately that the invariants listed in Theorem 7.4.1 form a SAGBI basis. This completes the proof of Theorem 7.4.1.   We will now give an example to illustrate the ideas in the above proof of Theorem 7.4.1. Example 7.4.9. We take p = 5 and m = 3 and consider invariants of multi5 degree (1, 1, 2). In symbols, we are considering F[3 V2 ]C (1,1,2) . We have the full polarization P : F[3 V2 ](1,1,2) → F[4 V2 ](1,1,1,1) . To avoid double subscripts, we will abuse notation by using the bases {x1 , y1 , x2 , y2 , x3 , y3 } of (3 V2 )∗ and {x1 , y1 , x2 , y2 , x3 , y3 , x4 , y4 } of (4 V2 )∗ . Here P(z) = z for z ∈ {x1 , y1 , x2 , y2 } and P(x3 ) = x3 + x4 and P(y3 ) = y3 + y4 . Accordingly, R(z) = z for z ∈ {x1 , y1 , x2 , y2 , x3 , y3 } and R(x4 ) = x3 and R(y4 ) = y3 . This full polarization equivariantly embeds F[3 V2 ](1,1,2) ∼ = V2 ⊗ V2 ⊗ 2 S (V2 ) ∼ = V2 ⊗ V2 ⊗ V3 into V2 ⊗ V2 ⊗ V2 ⊗ V2 . The image of this embedding is spanned by the 12 elements z1 z2 z2 where z1 ∈ {x1 , y1 }, z2 ∈ {x2 , y2 } and z3 ∈ {x3 x4 , x3 y4 + y3 x4 , y3 y4 }. By Lemma 7.4.5, we see that there are 5 partial Dyck paths of length d = 4 and height at most p − 2 = 3. The partial Dyck words corresponding to these 5 paths are Γ1 = RFRF, Γ2 = RRFF, Γ3 = RRRF, Γ4 = RRFR and Γ5 = RFRR. Furthermore, there is one word corresponding to an initially Dyck path of length d = 4 and escape height p − 1 = 4. This is the word Γ6 = RRRR. The paths corresponding to the first 5 words have finishing heights 0,0,2, 2 and 2 respectively. Of course, the final word, Γ6 , corresponds to a path which achieves height p−1 = 4. This means that ⊗4 V2 ∼ = 2 V1 ⊕3 V3 ⊕V5 . This decomposition can also be found using Lemma 7.4.4. Applying Λ, we get Λ(Γ1 ) = x1 y2 x3 y4 , Λ(Γ2 ) = x1 x2 y3 y4 , Λ(Γ3 ) = x1 x2 x3 y4 , Λ(Γ4 ) = x1 x2 y3 x4 , Λ(Γ5 ) = x1 y2 x3 x4 , and Λ(Γ6 ) = x1 x2 x3 x4 . We recognize these 6 monomials as the lead terms of the 6 invariants: F1 := θ(Γ1 ) = u12 u34 , F2 := θ(Γ2 ) = u14 u23 , F3 := θ(Γ3 ) = x1 x2 u34 ,

124

7 The Cyclic Group Cp

F4 := θ(Γ4 ) = x1 u23 x4 , F5 := θ(Γ5 ) = u12 x3 x4 , and F6 := θ(Γ6 ) = x1 x2 x3 x4 = − Tr(y1 y2 y3 y4 ). Note that u13 u24 = u12 u34 + u14 u23 , x2 u13 x4 = x1 u23 x4 + x3 u12 x4 and x1 x3 u24 = x1 x2 u34 + x1 x4 u23 are not required as basis elements. Restituting these 6 invariants we get R(F1 ) = 0, f2 := R(F2 ) = u13 u23 , R(F3 ) = 0, f4 := R(F4 ) = x1 u23 x3 , f5 := R(F5 ) = u12 x23 , and f6 := R(F6 ) = x1 x2 x23 = − Tr(y1 y2 y32 ). The two invariants F1 and F3 restitute to 0 since these two invariants do not lie in P(F[3 V2 ](1,1,2) ). Note that f4 = Δ2 ( 12 y1 u23 y3 ) and f5 = Δ2 ( 12 u12 y32 ) and thus f4 and f5 both lie inside copies of V3 . Also, f6 = Δ4 (−y1 y2 y32 ) spans the fixed line in a copy of V5 . Therefore, we see that {f2 , f4 , f5 , f6 } forms a basis for 2 C5 ∼ 5 ∼ F[3 V2 ]C = (V2 ⊗ V2 ⊗ V3 )C5 ∼ = (V1 ⊕ 2 V3 ⊕ V5 )C5 . (1,1,2) = (V2 ⊗ V2 ⊗ S (V2 ))

7.5 Integral Invariants In this section, we describe natural invariants in characteristic 0 which map to elements of F[Vn ]Cp under mod p reduction. Much of this discussion follows the paper of Shank [98]. Consider ⎛ ⎞ 1 0 0 ... 0 0 ⎜1 1 0 . . . 0 0⎟ ⎜ ⎟ ⎜0 1 1 . . . 0 0⎟ ⎜ ⎟ σ = ⎜. . . . . .⎟ ⎜ .. .. .. . . .. .. ⎟ ⎜ ⎟ ⎝0 0 0 . . . 1 0⎠ 0 0 0 ... 1 1 an n × n matrix with entries in the field of rational numbers Q. Then σ generates an infinite cyclic subgroup of SL(n, Q). We denote this subgroup by Z. Let Wn denote the n dimensional Q vector space on which σ and hence Z acts. Let {x1 , x2 , . . . , xn } denote the basis of Wn∗ dual to the standard basis.. We consider the ring of invariants Q[Wn ] = Q[x1 , . . . , xn ]Z . Note that since Z is infinite, we have no reason to expect that this ring of invariants is finitely generated. Surprisingly, it turns out to be so. Remark 7.5.1. By way of explaining this latter remark, consider the group H∼ = Z generated by ! 10 σ= 11 "

so that H=

! # 10 |i∈Z . i1

This group sits naturally in SL2 (C). The Zariski closure of H is the group " ! # 10 Ga (C) = |z∈C . z1

7.5 Integral Invariants

125

Here we use the notation Ga (C) to emphasize that this group is isomorphic to the group of complex numbers under addition. Let V be any representation of SL2 (C) and let C2 denote the standard 2-dimensional representation of SL2 (C) with basis {e1 , e2 }. Consider the algebra map ρ : C[V ⊕ C2 ] → C[V ] taking f to the function ρ(f ) given by ρ(f )(v) = f (v, e2 ). We note immediately that Ga (C) = {τ ∈ SL2 (C) | τ (e2 ) = e2 }. Roberts, [94], proved that the map ρ restricted as follows C[V ⊕ C2 ]SL2 (C) → C[V ]Ga (C) is an isomorphism of algebras. Since the former ring of invariants is known to be finitely generated by Hilbert’s famous result, so also is the latter. For the case needed here where G = SL2 (C), the finite generation of the ring of invariants was first proved by Gordan [48] or [49]. It is also true that Q[V ]Z = Q[V ]Ga (Q) . Famously, there is a unique n-dimensional irreducible representation of SL2 (C) which can be given by the natural action of SL2 (C) on % $ n−1 n−2 X ,X Y, . . . , Y n−1 , where (C2 )∗ has basis {X, Y }. This identification provides a map from SL2 (C) to GL(Wn ⊗ C). Under this map and with an appropriate choice of basis, the matrix ! 10 11 is carried to

⎛ ⎞ 1 0 0 ... 0 0 ⎜1 1 0 . . . 0 0⎟ ⎜ ⎟ ⎜0 1 1 . . . 0 0⎟ ⎜ ⎟ ⎜ .. .. .. . . .. .. ⎟ . ⎜. . . . . .⎟ ⎜ ⎟ ⎝0 0 0 . . . 1 0⎠ 0 0 0 ... 1 1

There are excellent references for this material: the primer of Kraft and Procesi [70], and the book of Procesi [90]. Given an element f of Q[Wn ]Z , we may clear denominators by multiplying by an appropriate integer m to obtain f  = mf ∈ Z[x1 , . . . , xn ]Z . Then reduction modulo p gives a map from Z[x1 , . . . , xn ]Z to Fp [x1 , . . . , xn ]Cp ⊆ F[x1 , . . . , xn ]Cp = F[Vn ]Cp . Any element of F[Vn ]Cp that may be constructed in this manner is called a integral invariant. Note the abuse of notation here in which {x1 , . . . , xn } is used to denote the dual bases over both Q and Fp and σ is used to represent a generator of both Z and Cp .

126

7 The Cyclic Group Cp

Classical invariant theorists have long recognized the difficulty of describing C[V ⊕C2 ]SL2 (C) . Even today, the description is complete for only a handful of representations. The material above indicates that describing the integral invariants of Fp [Vn ]Cp is equivalent. The invariant x1 ∈ Q[Vn ]Z is associated to the invariant x1 ∈ F[Vn ]Cp for every field F of characteristic p and representation Vn of Cp for n ≥ 1. Similarly, x22 − 2x1 x3 − x1 x2 ∈ Q[Wn ]Z is associated to an infinite family of invariants x22 − 2x1 x3 − x1 x2 ∈ F[Vn ]Cp for every field F of characteristic p and representation Vn of Cp for n ≥ 3. Shank, [97] and [98], conjectured that every ring of invariants of a Cp representation, F[Vn ]Cp , is generated by the three classes of invariants: norms, transfers and integral invariants. By norms here we mean only the norm N(xn ) of the distinguished variables xn ∈ Vn∗ . In order to discuss integral invariants comprehensively, we introduce the (non-finitely generated) ring Q[V∞ ] := Q[x1 , x2 , x3 , . . . ] over Q having infinitely many generators of degree 1. Of course, for any n ∈ N, we have the surjective map θ : Q[V∞ ] → Q[Vn ] given by

xi if i ≤ n, θ(xi ) = 0 if i > n. The integers,

Z, with generator σ, act on Q[V∞ ] via the left action generated by xi + xi−1 if i ≥ 2, σ(xi ) = As usual, we will use Δ to denote the operator x1 if i = 1. σ − Id. We will equip Q[V∞ ] with the graded lexicographic ordering with x1 < x2 < x3 < . . .. Lemma 7.5.2. Suppose n ≥ m > 1. Let β ∈ K[Vm−1 ] be a monomial and im−1 . Then βxm−1 and βxm are consecutive monomials write β := xi11 xi22 · · · xm−1 of K[Vn ] in the monomial ordering, i.e., if γ is a monomial of K[Vn ] with βxm−1 < γ ≤ βxm , then γ = βxm . Proof. Write γ = xj11 xj22 · · · xjnn and let a denote the least integer such that ia + δam−1 = ja . Since βxm−1 < γ ≤ βxm , we must have deg(γ) = deg(β) + 1 and thus βxm−1 < γ implies that ia + δam−1 > ja . If a < m − 1, then ia > ba would imply that βxm < γ, contradicting the hypotheses. If a > m − 1, then im−1 + 1 = jm−1 would imply that γ > βxm , again contradicting the hypothesis. Therefore, a = m − 1 and we have i1 = j1 , i2 = j2 , . . . , im−2 = jm−2 and im−1 + 1 = jm−1 . From, βxm−1 < γ we see jm−1 < im−1 + 1. Conversely, from γ ≤ βxm , we see that jm−1 ≥ im−1 . Therefore, jm−1 = im−1 . Finally, from γ ≤ βxm , we conclude that jm ≥ 1 and since deg(γ) = deg(β) + 1, we must have jm = 1   and γ = βxm .

7.5 Integral Invariants

127

Lemma 7.5.3. Let f be any non-zero polynomial in K[x1 , x2 , . . . , xn ]. Then LT(Δ(f )) < LT(f ). Proof. It is easy to see that LT(Δ(β)) < β for any monomial β. Applying this to each term in f we see that LT(Δ(f )) < LT(f ).   Theorem 7.5.4. Suppose that n ≥ m > 1 and that β is a monomial in / LT(K[x1 , x2 , . . . , xn ]Z ). K[x1 , x2 , . . . , xm−1 ]. Then βxm ∈ Proof. Suppose that f ∈ K[x1 , x2 , . . . , xn ] with LT(f ) = βxm . Write f = βxm + f  . Then LT(f  ) < βxm , i.e, by Lemma 7.5.2, LT(f  ) ≤ βxm−1 . Now Δ(f ) = βxm−1 +Δ(β)xm +Δ(β)xm−1 +Δ(f  ). We will show that LT(Δ(f )) = βxm−1 = 0 and thus f ∈ / K[x1 , x2 , . . . , xn ]Z . First, we see that LT(Δ(f  )) < LT(f  ) ≤ βxm−1 . Secondly, Δ(β) < β implies that Δ(β)xm−1 < βxm−1 . Finally, LT(Δ(β)) < β implies that LT(Δ(β)xm ) < βxm and thus by Lemma 7.5.2 we get LT(β)xm = LT(Δ(β)xm ) ≤ βxm−1 . Since xm does not divide βxm−1 , we see that LT(Δ(β)xm ) = βxm−1 and therefore, LT(Δ(β)xm ) < βxm−1 . Thus LT(Δ(f )) = βxm−1 = 0 as claimed.   In contrast to Theorem 7.5.4, we have the following. Theorem 7.5.5. Consider a monomial m −1 β = xa1 1 xa2 2 · · · xam−1 ∈ F[x1 , x2 , . . . , xm−1 ]. m Let am ≥ 2 and n ≥ 1+ =1 a (−1). Then there exists f ∈ K[x1 , x2 , . . . , xn ]Z with LT(f ) = βxamm .

Proof. Suppose first that either K has characteristic 0 or that am < p := characteristic(K). Then LT(Δ(βxamm )) = am βxm−1 xamm −1 . We define and note that

am −2 f1 := βxamm − am βxm−1 xm xm+1 am −1 . LT(Δ(f1 )) < am βxm−1 xm

For j ≥ 2, if Δ(fj−1 ) = 0, then we will define fj inductively from fj−1 as follows. Write LT(Δ(fj−1 )) = γxkr where k ≥ 1 and γ ∈ K[x1 , x2 , . . . , xr−1 ]. Then define xr+1 . fj := fj−1 − γxk−1 r

128

7 The Cyclic Group Cp

Thus LT(Δ(fj )) < γxkr = LT(Δ(fj−1 )) and therefore, the sequence LT(Δ(f1 )), LT(Δ(f2 )), . . . is strictly decreasing. However, this may not guarantee that the algorithm terminates since the number of variables involved is increasing at the same time. However, since LT(fj ) = βxamm for all j, all that remains to show is that the algorithm terminates with some fj ∈ K[x1 , x2 , . . . , xn ]Z . To see that the algorithm does indeed terminate and that the final invariant uses at most n variables, we introduce the concept of the weight of a monomial. A monomial λ = xb11 xb22 · · · xbss s has weight, wt(λ) := =1 b . It is easy to see that every monomial appearing in Δ(λ) has weight strictly smaller than wt(λ). Also, note that wt(γxk−1 xr+1 ) = wt(γxkr ) + 1. Thus every monomial occurring in each fj r am has weight  at most wt(βxm ). Furthermore, each of these monomials also has degree m a . Thus if =1 m λ = xb11 xb22 · · · xbss s m s with =1 a and =1 b ≤ m bs ≥ 1 appears in some fj , then =1 b = a . From this we see that the largest value of s that can occur  =1  mcorresponds to the monomial xt1 xn where t = ( m a ) − 1 and t + n = =1  =1 a . Thus m m m    a − ( a ) + 1 = 1 + a ( − 1). n= =1

=1

=1

In fact, it can be shown that this monomial always does occur with non-zero coefficient in the invariant f . Finally, we consider the case where am ≥ p := characteristic(K) > 0. Dividing am by p we may write am = qp + r where 0 ≤ r < p. By the above, there exists h ∈ F[x1 , x2 , . . . , xn ]Z with LT(h) = βxrm . Since LT(N(xm )) = xp ,   we see that N(xm )q h is invariant and has lead term βxamm , as required. Definition 7.5.6. The algorithm used in the above proof to construct an invariant with given lead term was introduced in [98] and is known as Shank’s algorithm. When we apply Shank’s algorithm to the lead term xi−1 2 , we produce invariants si−1 ∈ F[Vn ]Cp for each i, 3 ≤ i ≤ n. This family of invariants plays an important role in our analysis of the fraction fields associated to the rings of invariants, hence we will refer to them as the Shank invariants. The algorithm yields the following invariants:

7.5 Integral Invariants

129

i−3 i−3 i−2 si−1 = xi−2 2 v0 + x1 x2 v1 + · · · + x1 x2 vi−3 + x1 vi−2

=

i−1 

xji x2i−2−j vj

j=0

where v0 = x2 , and v1 = −(i−1)x3 . It is not hard to see that, for 1 ≤ j ≤ i−3, we have vj = αj,3 x3 + · · · + αj,j+2 xj+2 , while vi−2 = αi−2,2 x2 + αi−2,3 x3 + · · · + αi−2,i xi , for coefficients αj,k ∈ F. Here are some examples: s2 = x2 (x2 ) − x1 x2 + x1 (−2x3 ) s3 = x22 (x2 ) + x1 x2 (−3x3 ) + x21 (−x2 + 3x4 ) s4 = x32 (x2 ) + x1 x22 (−4x3 ) + x21 x2 (2x3 − 6x4 ) +x31 (−x2 − 2x3 − 6x4 − 8x5 ) Of course, as an alternative to the work of Shank, we could proceed to show that the requirement that Δ(si−1 ) = 0 gives us a system of equations in the unknowns αj,k which can be shown to have a (unique monic) solution over the integers. The following lemma ensures that we may isolate xi from the expression for si−1 once we have inverted x1 . Lemma 7.5.7. The coefficient of the term xi−2 1 xi in the integral invariant si−1 is invertible in Fp . Proof. The lemma follows through the calculation of the coefficients of the following monomials occurring in Δ(si−1 ): Monomial xi−2 1 xi−1 xi−3 1 x2 xi−2 .. .

Coefficient αi−1,i + αi−2,i−1 αi−2,i−1 + 2αi−3,i−2 .. .

xi−j−1 xj−1 1 2 xi−j .. .

αi−j,i−1−j + jαi−2−j,i−j .. .

x21 xi−4 2 x3

α2,4 + (i − 3)α1,3 ,

and now Shank’s algorithm tells us that α1,3 = −(i − 1). We may conclude that the coefficient of xi−2 1 xi is (−1)i (i − 3)!(i − 1), which is invertible for 3 ≤ i ≤ n ≤ p, as required.

 

130

7 The Cyclic Group Cp

7.6 Invariant Fraction Fields and Localized Invariants We have good descriptions of the fraction field of a ring of invariants since we may apply Galois theory. But how much information can we expect to obtain about the ring of invariants from its fraction field? This question was examined by Campbell and Chuai [16] and is recalled here. See also §4.3. We think of the fraction field as the localization of the ring away from the prime ideal (0). Our approach is to begin by considering a ring localized at a fixed point of the Cp -action. First, let us consider the case V = Vn over a field F of characteristic p. We have F[Vn ] = F[x1 , x2 , . . . , xn ] such that σ(xi ) = xi + xi−1 , subject to the convention x0 = 0. We use the Shank polynomials i−2 i si−1 = xi−1 + · · · + αi−2,2 xi−1 2 1 x2 + · · · + (−1) (i − 1)(i − 3)!x1 xi .

Since, by Lemma 7.5.7, the coefficient of xi−2 1 xi is invertible in F, this equation can be solved for xi in the localized ring F[Vn ]x1 . For example, we have s2 = x22 − x1 x2 − 2x1 x3 , which we solve for x3 =

−1 (s2 − x22 + x1 x2 ). 2x1

Inductively, we may assume that x3 , . . . , xn can be written as functions of x±1 1 , x2 and the invariant rational functions si−1 , 3 ≤ i ≤ n. That is, we have proved Proposition 7.6.1. & & 3 ≤ i ≤ n]. F[Vn ]x1 = F[x±1 1 , x2 , si−1 Consequently, & p p−1 ±1 p x2 , si−1 & 3 ≤ i ≤ n] F[Vn ]C x1 = F[x1 , N (x2 ) = x2 − x1 and

& F(Vn )Cp = F(x1 , N (x2 ), si−1 & 3 ≤ i ≤ n).

The proposition reveals clearly the limitations of this method. We note that the invariant fraction field is generated by dim(V ) elements. Further, the ring of invariants localized at the single invariant x1 is also generated by dim(V ) many elements. That is, as soon as just one suitable element is inverted, the complex and subtle structure of the ring of invariants collapses to this relatively simple structure: no traces are needed to generate the localized rings and all of the generators needed are of degree p or less. In the event V has more than one non-trivial summand, say

7.6 Invariant Fraction Fields and Localized Invariants

131

V n ⊕ Vm ⊂ V with n, m > 1, we need a particular kind of integral invariant, the degree 2 determinant invariants. Let {x1 , x2 , . . . , xn } be a triangular basis for Vn∗ with xn distinguished. Similarly let {y1 , y2 , . . . , ym } denote a triangular bases for Vm∗ with ym distinguished. Then u = uVn ,Vm = x1 y2 − x2 y1 is invariant, and there is one such an invariant for every pair of non-trivial summands in the indecomposable decomposition of V . There is one more family of invariants that are needed, again associated to a representation with more than one non-trivial summand, say Vn ⊕ Vm ⊂ V with bases as above and n ≤ m. Suppose f = f (x1 , . . . , xn ) ∈ F[Vn ]Cp has degree d. One of the polarizations of f , fd−1,1 , has multi-degree (d − 1, 1) in Vn ⊕ Vn . We consider the second copy of Vn here as being embedded as a Cp -submodule of Vm . Thus we may interpret fd−1,1 as a bi-homogeneous polynomial of degree d − 1 in the variables x1 , x2 , . . . , xn and degree 1 in y1 , y2 , . . . , yn . With this interpretation, we denote fd−1,1 by f1 . We are interested only in applying this construction to the Shank invariants. In the three examples above, these are (s2,n,m )1 = −x2 y1 + 2x2 y2 − 2x3 y1 + x1 (−y2 − 2y3 ) (s3,n,m )1 = 6x4 x1 y1 − 3x3 x2 y1 − 3x3 x1 y2 + 3x22 y2 −3x2 x1 y3 − 2x2 x1 y1 + x21 (−y2 + 3y4 ) (s4,n,m )1 = −24x5 x21 y1 + 16x4 x2 x1 y1 + 8x4 x21 y2 −18x4 x21 y1 − 4x3 x22 y1 − 8x3 x2 x1 y2 + 4x3 x2 x1 y1 +2x3 x21 y2 − 6x3 x21 y1 + 4x32 y2 − 4x22 x1 y3 +8x2 x21 y4 + 2x2 x21 y3 − 3x2 x21 y1 − x31 (−y2 − 2y3 − 6y4 + 8y5 ), Note the abuse of notation here: s2,n,m = s2,Vn ,Vm . Note as well that si,n,m is only defined for i ≤ min(n, m). Again, it is critical to note that the coefficient of xi−1 1 yi in (si,n,m )1 is invertible in F. This follows from Lemma 7.5.7. Decompose V as V = k1 V1 ⊕ k2 V2 ⊕ · · · ⊕ kp Vp and choose a triangular basis for V ∗ {xi,j,n | 1 ≤ i ≤ n, 1 ≤ j ≤ kn , 1 ≤ n ≤ p} with the property that σ(xi,j,n ) = xi,j,n + xi−1,j,n , subject to the convention that x0,j,n = 0. Thus F[V ] = F[xi,j,n | 1 ≤ i ≤ n, 1 ≤ j ≤ kn , 1 ≤ n ≤ p] Now we choose the largest m for which km > 0 and fix a particular copy ∗ V,m of Vm . Let {x1 , x2 , . . . , xm } denote the usual basis for V,m . Set x1 = x1,,m and x2 = x2,,m . Then σ(x1 ) = x1 and that σ(x2 ) = x2 + x1 .

132

7 The Cyclic Group Cp

For each other distinct non-trivial summand Vn , then, we have n ≤ m and we choose the usual basis {y1 , . . . , yn }, that is, we have σ(yi ) = yi + yi−1 with y0 = 0. We consider the polarized Shank invariants (si−1,n,m )1 for 3 ≤ i ≤ n − 1, each of which has a term xi−2 1 yi which occurs with coefficient (−1)i (i − 1)(i − 3)!, invertible in F. As above, we may inductively solve for yi , 3 ≤ i ≤ n, in the localized ring F[V ]x1 . Finally, we consider the determinant invariant un,m = uVn ,Vm = x1 y2 − x2 y1 and note that we have y2 =

1 (un,m + x2 y1 ). x1

Hence we have proved the following Theorem 7.6.2. Let V = k2 V2 ⊕ · · · ⊕ kp Vp be any reduced representation of Cp . Choose a triangular basis {xi,j,n | 1 ≤ i ≤ n} for the jth copy of Vn∗ for j = 1, 2, . . . , kn for n = 2, 3, . . . p. Fix a choice x1 = x1,,m and x2 = x2,,m for some  and m the largest m for which km > 0. Then & & 1 ≤ j ≤ kn , 2 ≤ n ≤ p] F[V ]x1 = F[x±1 1 , x2 , x1,j,n , un,m , (si,n,m )1 and & & 1 ≤ j ≤ kn , 2 ≤ n ≤ p) . F(V )Cp = F(x±1 1 , N (x2 ), x1,j,n , un,m , (si,n,m )1

7.7 Noether Number for Cp In this section, we will present a theorem of Fleischmann, Sezer, Shank and Woodcock which gives exact values for the Noether number of the ring of invariants of any representation of Cp . Here we will follow their proof closely. However, we will change it in order to make it self-contained, The proof given below does not rely on results from [100]. Recall that we say that V is reduced if V does not contain a copy of V1 as a summand. Theorem 7.7.1. Let V be a non-trivial reduced representation of Cp , the cyclic group of order p defined over a field F of characteristic p > 0. Let s be the maximum dimension of an indecomposable summand of V . (Thus 2 ≤ s ≤ p.) Then ⎧ p, ⎪ ⎪ ⎪ ⎨(p − 1) dim(V Cp ), β(V, Cp ) = ⎪(p − 1) dim(V Cp ) + 1, ⎪ ⎪ ⎩ (p − 1) dim(V Cp ) + p − 2,

if if if if

V ∼ = 2 V2 ; = V2 or V ∼ V ∼ = m V2 for m ≥ 3; s = 3; s ≥ 4.

7.7 Noether Number for Cp

133

Note that dim V Cp is the number of indecomposable summands in V . If s = 2, then the result follows from the Theorem 7.4.1 (and Remark 7.4.2). Thus we suppose that s ≥ 3. We begin the proof of Theorem 7.7.1 by developing lower bounds for β(V, Cp ). To do this, we consider the two Cp -representations U = V4 ⊕ m V2 and W = V3 ⊕ m V2 . Choose a triangular basis {w0 , z0 , y0 , x0 } for the submodule V4∗ of U ∗ and triangular bases {yi , xi } (i = 1, 2, . . . , m) for the duals of the m summands isomorphic to V2 . Thus Δ(w0 ) = z0 , Δ(z0 ) = y0 , Δ(yi ) = xi and Δ(xi ) = 0 for i = 0, 1, . . . , m. Then {w0 , z0 , y0 , x0 }  {yi , xi | i = 1, 2, . . . , m} is a basis for U ∗ . Viewing W as a Cp submodule of U we may take {z0 , y0 , x0 }  {yi , xi | i = 1, 2, . . . , m} as a basis for W ∗ . Lemma 7.7.2. Take U = V4 ⊕ m V2 and W = V3 ⊕ m V2 as above. Then TrCp ((w0 y1 y2 · · · ym )p−1 z0p−2 ) is an indecomposable element of F[U ]Cp and TrCp ((z0 y1 y2 · · · ym )p−1 y0 ) is an indecomposable element of F[W ]Cp . Proof. We consider U first. Put Fm := TrCp ((w0 y1 y2 · · · ym )p−1 z0p−2 ). We proceed by induction on m. For m = 0, we are concerned with F[V4 ]Cp . Shank [98] found a set of generators for F[V4 ]Cp . We describe this computation at length in Chapter 13. We will use the graded reverse lexicographic order on F[V4 ] determined by w0 > z0 > y0 > x0 . Using Lemma 9.0.2, we can show that LT(F0 ) = −z02p−3 . The (non-minimal) generating set for F[V4 ]Cp given by Shank includes invariants whose lead term is of the form z0i precisely for the values i = p − 1, p, . . . , 2p − 3 (including F0 ). Furthermore, the only invariant among the generators whose lead term is divisible by w0 is N(w0 ) with lead term w0p . This shows that the largest (with respect to the graded reverse lexicographic order) monomial in LT(F[V4 ]Cp )d is strictly smaller than z0d for all d = 1, 2, . . . , p − 2. In particular, there can be no (sequence of) tˆete-a-tˆetes among the generators yielding invariants with lead term z0d for d ≤ p − 1. This shows that F0 is indecomposable since its lead term cannot be expressed as a product of lead terms of lower degree invariants. The reader may find a more detailed version of this argument in [96]. For the general case m ≥ 1, we assume the induction hypothesis that Fm−1 is a non-zero indecomposable element of F[V4 ⊕ (m − 1) V2 ]Cp . We consider the Cp submodule U  = V4 ⊕ (m − 1) V2 ⊕ V1 of U whose dual has basis {w0 , z0 , y0 , x0 , y1 , x1 , . . . , ym−1 , xm−1 , xm }. The inclusion of U  into U induces a surjective, multi-degree preserving, Cp -equivariant algebra map π : F[U ] → F[U  ] determined by π(ym ) = xm , π(ym−1 ) = xm−1 and fixing all the other variables. Now  π(Fm ) = π( σ((z0 y1 y2 · · · ym−1 ym )p−1 y0p−2 )) σ∈Cp

=



π(σ((z0 y1 y2 · · · ym−1 ym )p−1 y0p−2 ))

σ∈Cp

=



σ∈Cp

σ(π((z0 y1 y2 · · · ym−1 ym )p−1 y0p−2 ))

134

7 The Cyclic Group Cp

=

 σ∈Cp

p−1 = xm

σ((z0 y1 y2 · · · ym−1 xm )p−1 y0p−2 ) 

σ((z0 y1 y2 · · · ym−1 )p−1 y0p−2 )

σ∈Cp

= =

p−1 TrCp ((z0 y1 y2 xm p−1 xm Fm−1 .

· · · ym−1 )p−1 y0p−2 )

Note that since Fm−1 = 0, this shows that Fm = 0. Assume byway of contradiction that Fm is decomposable in F[U ]Cp and write Fm = ri=1 fi gi for C some multi-homogeneous invariants fi , gi ∈ F[U ]+p . Thus each of the products fi gi has multi-degree (2p − 3, p − 1, p − 1, . . . , p − 1). Let xami denote the largest power of xm which divides fi for i = 1, 2, . . . , r. Similarly, let xbmi denote the largest power of xm which divides gi . Thus ai +bi = p−1 for all i = 1, 2,r . . . , r. p−1 have π(F ) = x Therefore, writing fi = xami fi and gi = xbmi gi we m m i=1 fi gi r   with fi and gi in F[U  ]Cp . Hence Fm−1 := f g . Since F is indem−1 i=1 i i composable, one of the fi or gi , say fi , must be a non-zero constant, c. p−1 Thus π(fi ) = cxm and therefore, fi has multi-degree (0, 0, 0, . . . , 0, p − 1). Cp p p−1 p−1 = F[ym − xm ym , xm ]p−1 = spanF {xm }. But then Hence fi ∈ F[ym , xm ]p−1  π(fi ) = 0 contradicting the fact that fi = c. This contradiction shows that Fm is indecomposable in F[U ]Cp . The proof that TrCp ((z0 y1 y2 · · · ym )p−1 y0 ) is indecomposable in F[W ]Cp is similar. We proceed by induction as above using Lemma 7.7.2 and using Theorem 4.10.1 to handle the base case.   Corollary 7.7.3. Let V be a non-trivial reduced representation of Cp . Let s be the maximum dimension of an indecomposable summand of V . Suppose s ≥ 3. Then

(p − 1) dim(V Cp ) + 1, if s = 3; β(V, Cp ) ≥ Cp (p − 1) dim(V ) + p − 2, if s ≥ 4. Proof. We treat the case s = 4 first. Put m := dim(V Cp ) − 1 and U := V4 ⊕ m V2 . Decompose V as V = Vr0 ⊕ Vr1 ⊕ · · · ⊕ Vrm where r0 ≥ 4 and ri ≥ 2 for i = 1, 2, . . . , m. For each i = 0, 1, 2, . . . , m, choose a distinguished variable Xi ∈ Vr∗i . We determine a Cp -equivariant surjection φ : V ∗ → U ∗ by requiring that φ(X0 ) = w0 and φ(Xi ) = yi for i = 1, 2, . . . , m. This induces a Cp -equivariant surjection, which we also call φ mapping F[V ] onto F[U ]. Thus φ(Δj (Xi )) = Δj (φ(Xi )) for all i and j. Restricting to invariants we get an algebra map φ : F[V ]Cp → F[U ]Cp . Since φ is an algebra homomorphism and since φ(TrCp ((X0 X1 X2 · · · Xm )p−1 Δ(X0 )p−2 )) = TrCp ((w0 y1 y2 · · · ym )p−1 z0p−2 ) is indecomposable in F[U ]Cp , it follows that

7.7 Noether Number for Cp

135

TrCp ((X0 X1 X2 · · · Xm )p−1 Δ(X0 )p−2 ) is indecomposable in F[V ]Cp . Therefore, β(V, Cp ) ≥ (m + 1)(p − 1) + p − 2. The case s = 3 is proved similarly.   Our next goal is, of course, to show that any invariant whose degree exceeds the bound given in Corollary 7.7.3 must be decomposable. Accordingly, we consider a degree a with a ≥ (m+2)(p−1). By the discussion preceding Theorem 7.3.2, we see that every f ∈ F[V ]G may be written as f = f  + f  where f  ∈ F[V ]a and f  ∈ F[V ]a . Since f  ∈ F[V ]a , we may write  Cp Cp f  = m+1 i=1 fi N (zi ) with fi ∈ F[V ]a−p . Therefore, f is indecomposable if and only if f  is indecomposable. Thus we concentrate our attention on f  . Since F[V ]a is free, (F[V ]a )Cp = TrCp (F[V ]a ). Thus to prove Theorem 7.7.1, it suffices to show that every homogeneous invariant of the form TrCp (h) of degree at least (m + 2)(p − 1) is decomposable. In order to study the image of the transfer, we consider F[V ] as an F[V ]Cp -module. Since F[V ] is integral over F[V ]Cp , we know that F[V ] is a finitely generated F[V ]Cp -module, i.e., there exist h1 , h2 , . . . , ht ∈ F[V ] such that F[V ] = F[V ]Cp h1 + F[V ]Cp h2 + · · · + F[V ]Cp ht . By the graded Nakayama lemma 2.10.1, we may find a homogeneous minimal such set of generators h1 , h2 , . . . , ht by lifting back to F[V ] any homogeneous basis for the maximal homogeneous ideal in the ring of coinvariants F[V ]Cp = F[V ]/J where J is the Hilbert ideal. We define γ(V ) = max{d | (F[V ]Cp )d = {0}}. Hence for a homogeneous minimal set of generators, h1 , h2 , . . . , ht , we have γ(V ) = max{deg(hi ) | 1 ≤ i ≤ t}, Thus γ is the degree of the Hilbert series H(F[V ]Cp , λ) (which is in fact a polynomial). Given any homogeneous h ∈ F[V ], we may write h = f1 h1 +f2 h2 +· · ·+ft ht with f1 , f2 , . . . , ft ∈ F[V ]Cp and deg(h) = deg(fi ) + deg(hi ) for i = 1, 2, . . . , t. Therefore, TrCp (h) = f1 TrCP (h1 )+f2 TrCP (h2 )+· · ·+ft TrCP (ht ). This shows that if deg(h) > γ(V ), then TrCp (h) is a decomposable invariant. Our next step is to get upper bounds (which will turn out to be sharp) for γ(V ). As above, we let s denote the maximum dimension of an indecomposable summand of V . We will show that

(p − 1) dim(V Cp ) + 1, if s = 3; γ(V ) = Cp (p − 1) dim(V ) + p − 2, if s ≥ 4. As we noted above, γ(V ) = deg(H(F[V ]Cp , λ). Thus γ(V ) = deg(H(F[V ], λ) − deg(H(J, λ) = deg(H(F[V ], λ) − deg(H(LT(J), λ) = deg(H(F[V ]/ LT(J), λ). This is true regardless of which monomial order is used to determine lead terms.

136

7 The Cyclic Group Cp

Thus γ(V ) is the highest degree of a monomial not lying in LT(J). We will use this characterization to determine γ(V ). We consider the case s = 3 first. For this case, we may write V = m1 V2 ⊕ m2 V3 where m2 ≥ 1. We choose distinguished variables y1 , y2 , . . . , ym1 and zm1 +1 , zm1 +2 , . . . , zm1 +m2 . We define yi = Δ(zi ) for m1 + 1 ≤ i ≤ m1 + m2 and xi = Δ(yi ) for 1 ≤ i ≤ m1 + m2 . Then we have F[V ] = F[zm1 +1 , zm1 +2 , . . . , zm1 +m2 , y1 , y2 , . . . , ym1 +m2 , x1 , x2 , . . . , xm1 +m2 ]. We fix a graded reverse lexicographic order with respect to these variables and satisfying y1 < zi for m1 + 1 ≤ i ≤ m1 + m2 and xi < yi for all 1 ≤ i ≤ m1 + m2 . It is easily verified that di = yi2 − 2xi zi − xi yi is invariant with LT(di ) = yi2 for all m1 + 1 ≤ i ≤ m1 + m2 . Similarly, the polarizations of di given by dij = yi yj − xi zj − zj xi − xi yj are also invariant and satisfy LT(dij ) = yi yj for all m1 + 1 ≤ i < j ≤ m1 + m2 . The linear monomials x1 , x2 , . . . , xm1 +m2 are all invariant. Finally, zip = LT(NCp (zi )) ∈ LT(J) for all 1 ≤ i ≤ m1 and yip = LT(NCp (yi ) ∈ LT(J) for all m1 + 1 ≤ i ≤ m1 + m2 . From this it is clear that a monomial not lying in LT(J) cannot be divisible by any xi , nor by any yi yj with i, j ≥ m1 + 1, nor by any zip , nor by any yip . This shows that the highest degree monomials which may lie outside LT(J) are those of the form (y1 y2 · · · ym1 zm1 +1 · · · zm1 +m2 )p−1 yj with j > m1 . Therefore, γ(m1 V2 ⊕m2 V3 ) ≤ (m1 +m2 )(p−1)+1 = (p−1) dim(V Cp )+1 which completes the proof for the case s = 3. The proof for the case s ≥ 4 is similar. In this case, we will need two small technical lemmas in order to bound γ(V ). Before stating and proving these lemmas, we fix some notation. As above, we write V = Vd1 ⊕ Vd2 ⊕ · · · ⊕ Vdm and we choose a distinguished variable zi in each Vd∗i . We then consider the basis B := {zij | 1 ≤ i ≤ m, 0 ≤ j < di } of V ∗ where zij := Δj (zi ) for all j = 0, 1, . . . , di − 1 and i = 1, 2, . . . , m. We also define B  := {zij | 1 ≤ i ≤ m, 1 ≤ j < di } = B \ {z1 , z2 , . . . , zm } and the polynomial ring R := F[B  ] = F[zij | 1 ≤ i ≤ m, 1 ≤ j < di ]. Let α be any monomial of degree p in R. Write α = u1 u2 . . . up−1 and define α = w1 w2 . . . wp−1 to be the monomial of degree p − 1 in F[V ] where ws ∈ B and Δ(ws ) = us for s = 1, 2, . . . , p − 1. We fix a graded reverse lexicographic order on F[V ] (and R) with zij < zik whenever j > k. Let S be any subset of  {1, 2, . . . , p − 1}. We will denote {1, 2, . . . , p − 1} \ S by S  . We define XS := s ∈ Sws and XS  := s ∈ S  ws so that for all   subsets S, we have XS XS = α . Fix a generator σ of Cp and consider the function F :=

p−1 p−1  

(wj − σ t (wj ))

t=0 j=1

=

 p−1  τ ∈Cp j=1

(wj − τ (wj )) .

7.7 Noether Number for Cp

137

Lemma 7.7.4. 

F =

(−1)|S| XS  TrCp (XS ) .

S⊆{1,2,...,p−1}

Proof. For each t we have p−1 



(wj − σ t (wj )) =

j=1

(−1)|S| XS  σ t (XS )

S⊆{1,2,...,p−1}

Thus F =

p−1 p−1   (wj − σ t (wj )) t=0 j=1

=

p−1 



(−1)|S| XS  σ t (XS )

t=0 S⊆{1,2,...,p−1}



=

(−1)|S| XS 

S⊆{1,2,...,p−1}



=

p−1 

σ t (XS )

t=0

(−1)|S| XS  TrCp (XS )

S⊆{1,2,...,p−1}

  The function F defined above satisfies the following. Lemma 7.7.5. LT(F ) = −α . In particular, α ∈ LT(J).

  Proof. Since σ t (wj ) = wj + tΔ(wj ) + 2t Δ2 (wj ) + · · · = wj + tuj + . . . we see that LT(wj − σ t (wj )) = −tuj for all t ≥ 1 and all j. Therefore, if 1 ≤ t ≤ p − 1 p−1 p−1 p−1 we have LT( j=1 (wj − σ t (wj )) = j=1 LT(wj − σ t (wj )) = j=1 −tuj = (−t)p−1 α = α. For t = 0, we have wj − σ 0 (wj ) = 0. Therefore, F is the sum of p − 1 non-zero summands, each of which has lead term α. Thus LT(F ) = (p − 1)α = −α.   Note that zip = LT(NCp (zi )) ∈ LT(J) for all 1 ≤ i ≤ m. Thus we have shown that the monomials of largest degree which may lie outside LT(J) are those monomials of the form (z1 z2 · · · zm )p−1 α where α = u1 u2 · · · up−2 is some monomial in R of degree p − 2. Thus γ(V ) ≤ m(p − 1) + p − 2. This completes the proof of Theorem 7.7.1.

138

7 The Cyclic Group Cp

7.8 Hilbert Functions Hughes and Kemper [54] provide an explicit formula for the calculation of the Hilbert series of an arbitrary representation of Cp in characteristic p as follows. We have that V = Vn1 +1 ⊕ Vn2 +1 ⊕ · · · ⊕ Vnk +1 where each Vn +1 is an indecomposable Cp -module of dimension n + 1, 0 ≤ n < p. We define a function DV (ξ, t) ∈ C[t] as ⎛ ⎞ ni k pn p   1 − ξ t ⎝ DV (ξ, t) = (1 − ξ n −2j )−1 ⎠ . 1 − tp j=0

=1

Theorem 7.8.1. Let M(2p) ⊂ C denote the set of 2p-th roots of unity. Then, for V and DV as just defined, the Hilbert series of F[V ]Cp is given by H(F[V ]Cp , t) =

 ξ∈M(2p)

1+ξ DV (ξ, t). 2p

If all of the n are even, then the formula simplifies to H(F[V ]Cp , t) =

1 p

n k   

(1 − ξ n −2j t)−1 .

ξ∈M(p) =1 j=0

Furthermore, the multiplicity of Vn in F[V ] as a Cp -module is given by  ξ∈M(2p)

ξ − ξ −1 DV (ξ, t), 2p

if n < p,

while the multiplicity of Vp in F[V ] as a Cp -module is given by  ξ∈M(2p)

1+ξ p ξ DV (ξ, t) . 2p

Remark 7.8.2. Suppose that V is a sum of odd dimensional indecomposables k := ⊕Vni +1 Vi=1

where all

ni are even.

Fix a primitive p-th root of unity ξ ∈ C. Define the matrices M = diag(ξ n , ξ n −2 , . . . , ξ, 1, ξ −1 , . . . , ξ (n −2) , ξ −n ) ∈ GLn +1 (C) and set M = diag(M1 , M2 , . . . , Mk ) ∈ GLdim(V ) (C).

7.8 Hilbert Functions

139

We obtain a non-modular representation of Cp by mapping a generator to M . We may use Molien’s theorem Theorem 3.7 to determine the Hilbert series of C[M ]Cp . It is a remarkable fact that this Hilbert series is the same as the Hilbert series of F[V ]Cp . Both Almkvist [2] and Hughes and Kemper [54] believe this to be a “combinatorial accident”.

8 Polynomial Invariant Rings

In this chapter, we study representations of groups which have polynomial rings of invariants. In characteristic 0, this happens if and only if the group is a reflection group. Recall 1.5.1 Definition 8.0.1. Suppose V is a representation of G defined over a field K of any characteristic. We say that σ ∈ G is a reflection on V if the dimension of V σ is dim(V ) − 1, that is, e = σ fixes a hyperplane pointwise. A reflection σ ∈ GL(V ) is said to be a transvection if it is not diagonalizable. If it is diagonalizable, it is said to be a homology. Building on work of Shephard and Todd [101], and Chevalley [22], Serre proved Theorem 8.0.2. If K[V ]G is a polynomial ring, then G is generated by reflections. Lemma 8.0.3. Suppose V is a representation of G defined over a field F of characteristic p > 0. If σ ∈ G is a reflection whose order is pr , then σ has order p. Proof. Choose any element v ∈ V lying outside of the hyperplane V σ . Then σ(v) = λv + w for some λ ∈ F and w ∈ V σ . Now σ  (v) = λv + (λ−1 + λ−2 + r r r r · · · + 1)w. Therefore, σ p (v) = λp v + ((λp −1 + λp −2 + · · · + 1)w = v. In r particular, λp = 1 and so as a (pr )th -root of unity, λ = 1. Thus σp (v) = v.   Therefore, σ p fixes all of V pointwise as claimed. Definition 8.0.4. Let P be a p-subgroup of GL(V ) where V is an n dimensional vector space over the field F of characteristic p. We say that P is a Nakajima group (with respect to B) if there exists an ordered basis B = {x1 , x2 , . . . , xn } of V ∗ such that 1. σ(xi ) −xi ∈ span{x1 , x2 , . . . , xi−1 } for all i = 1, 2, . . . , n and for all σ ∈ P , i.e., in the basis B, P is an upper triangular group, and H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, 141 Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 8, © Springer-Verlag Berlin Heidelberg 2011

142

8 Polynomial Invariant Rings

2. P = Pn Pn−1 · · · P1 := {σn σn−1 · · · σ1 | σi ∈ Pi } where Pi := {σ ∈ P | σ(xj ) = xj for all j = i}. If P is a Nakajima group, we call the basis B a Nakajima basis. An example of a Nakajima basis is given in Example 8.0.8. Lemma 8.0.5. For i < j, Pi normalizes Pj , that is, Pi Pj = Pj Pi .

 

From this it follows that the set Pn Pn−1 · · · P2 P1 is in fact a subgroup of P . It is also clear that Pi ∩ Pj = {e} for all i = j. Lemma 8.0.6. Suppose that B = {x1 , x2 , . . . , xn } is a basis of V ∗ and that P is a finite subgroup of GL(V ). As above, let Pk = {σ ∈ P | σ(xj ) = xj for all j = k}. Then every element γ ∈ Pn Pn−1 · · · P1 has a unique expression of the form γ = σn σn−1 · · · σ1 with σi ∈ Pi for i = 1, 2, . . . , n. Proof. We proceed by induction on n with the case n = 1 being trivial. Suppose the result holds for n − 1 and assume σn σn−1 · · · σ1 = τn τn−1 · · · τ1 where σi , τi ∈ Pi for i = 1, 2, . . . , n. Let τ denote the element τn−1 σn = −1 τn−1 τn−2 · · · τ1 σ1−1 σ2−1 · · · σn−1 . Then  if i < n; τn−1 σn (xi ) = xi τ (xi ) = −1 (xi ) = xn , if i = n. τn−1 τn−2 · · · τ1 σ1−1 σ2−1 · · · σn−1 Thus τ ∈ P1 ∩ P2 ∩ · · · ∩ Pn = {e} and therefore τn = σn . This implies σn−1 σn−2 · · · σ1 = τn−1 τn−2 · · · τ1 . Applying the induction hypothesis yields   τi = σi for all i = 1, 2, . . . , n − 1. Theorem 8.0.7. Let P be a p-subgroup of GL(V ) and suppose that B = {x1 , x2 , . . . , xn } is an ordered basis of V ∗ with respect to which P is an upper triangular subgroup of GL(V ∗ ). Then G is a Nakajima group with Nakajima basis B if and only if F[V ]P = F[N1 , N2 , . . . , Nn ] where Ni = NP Px (xi ). i

Proof. First we assume that B is a Nakajima basis for P . Since P = Pn Pn−1 · · · P1 , Lemma 8.0.6 implies that |P | =

n 

|Pk |.

k=1

Since σn σn−1 · · · σ1 (xk ) = σk (xk ) where σi ∈ Pi for i = 1, 2, . . . , n, we see that P · xk = Pk · xk . Furthermore, if σk , τk ∈ Pk are such that σk (xk ) = τk (xk ), then σk−1 τk (xk ) = xk and thus σk−1 τk = P1 ∩ P2 ∩ · · · ∩ Pn = {e}. This shows that |Pk · xk | = |Pk |. Therefore,

8 Polynomial Invariant Rings

143

n n deg Nk = |P ·xk | = |Pk ·xk | = |Pk | and thus k=1 deg(Nk ) = k=1 |Pk | = |P |. By Proposition 4.0.3, N1 , N2 , . . . , Nn is a homogeneous system of parameters for F[V ]P . Therefore, by Corollary 3.1.6, F[V ]P = F[N1 , N2 , . . . , Nn ]. The proof of the opposite direction follows immediately from Theorem 8.0.11 below.   Example 8.0.8. Define ⎛

10 ⎜0 1 ⎜ ⎜0 0 ⎜ σ(a, b, c, d) = ⎜ ⎜a 0 ⎜0 b ⎜ ⎝0 0 dd

0 0 1 0 0 c d

⎞ 0000 0 0 0 0⎟ ⎟ 0 0 0 0⎟ ⎟ 1 0 0 0⎟ ⎟ 0 1 0 0⎟ ⎟ 0 0 1 0⎠ 0001

and let G denote the subgroup of GL(V ) given by these matrices as a, b, c and d vary over Fp . We first saw this group in §1.1.1. The action of σ(a, b, c, d) on V ∗ is given by the matrix ⎛ ⎞ 1 0 0 −a 0 0 −d ⎜0 1 0 0 −b 0 −d⎟ ⎜ ⎟ ⎜0 0 1 0 0 −c −d⎟ ⎜ ⎟ ⎜0 0 0 1 0 0 0 ⎟ ⎜ ⎟ ⎜0 0 0 0 1 0 0 ⎟ ⎜ ⎟ ⎝0 0 0 0 0 1 0 ⎠ 000 0 0 0 1 with respect to the basis B dual to the standard basis. It is clear that P1 = P2 = P3 = {e} while P4 = {σ(a, 0, 0, 0) | a ∈ Fp } , P5 = {σ(0, b, 0, 0) | b ∈ Fp } , P6 = {σ(0, 0, c, 0) | c ∈ Fp } and P7 = {σ(0, 0, 0, d) | d ∈ Fp } . It is also clear that the Abelian group G = P7 P6 · · · P1 , and so B is a Nakajima basis for G and thus F[V ]G is a polynomial ring. By way of contrast in this next example, we return to the group H defined in §1.1.1 as the transpose of the G of the previous example acting on V . Example 8.0.9. The transpose of σ(a, b, c, d) has the form

144

8 Polynomial Invariant Rings

⎛

⎞ 100a00d ⎜0 1 0 0 b 0 d⎟ ⎜ ⎟ ⎜0 0 1 0 0 c d⎟ ⎜ ⎟ ⎟ σ(a, b, c, d) = ⎜ ⎜0 0 0 1 0 0 0⎟ . ⎜0 0 0 0 1 0 0⎟ ⎜ ⎟ ⎝0 0 0 0 0 1 0⎠ 0000001 We consider the group H ⊂ GL(V ) given by these latter matrices as a, b, c and d vary over Fp . Following our usual convention, we rewrite the action of H on V in lower triangular form. The matrix just given is written with respect to the standard basis {e1 , e2 , . . . , e7 }. Re-ordering this basis as {e4 , e5 , e6 , e7 , e1 , e2 , e3 } and re-writing the matrix accordingly we obtain ⎛ ⎞ 1000000 ⎜0 1 0 0 0 0 0⎟ ⎜ ⎟ ⎜0 0 1 0 0 0 0⎟ ⎜ ⎟ ⎟ σ(a, b, c, d) = ⎜ ⎜0 0 0 1 0 0 0⎟ . ⎜a 0 0 d 1 0 0⎟ ⎜ ⎟ ⎝0 b 0 d 0 1 0⎠ 00cd001 We will show that F[V ]H is not Cohen-Macaulay. We define α = σ(1, 0, 0, 0), β = σ(0, 1, 0, 0), γ = σ(0, 0, 1, 0) and δ = σ(0, 0, 0, 1). Let K be the subgroup generated by α, β and γ. We label the basis dual to the basis {e5 , e6 , e7 , e1 , e2 , e3 , e4 } as {x1 , x2 , x3 , x, y1 , y2 , y3 } so that (V ∗ )G = span {x1 , x2 , x3 , x}. Define Ni (t) = tp − xip−1 t. By Nakajima’s Theorem 8.0.7, we have F[V ]K = F[x1 , x2 , x3 , x, N1 (y1 ), N2 (y2 ), N3 (y3 )]. Now define Δ = δ − 1. The rest of this example flows from Example 4.0.4. We calculate Δ(Ni (yi )) = x(xp−1 − xip−1 ) = Ni (x) and hence uij := (xp−1 − xip−1 )Nj − (xp−1 − xjp−1 )Ni is H-invariant for 1 ≤ i < j ≤ 3. It is easy to verify that x1 u23 − x2 u13 + x3 u12 = 0 and that {x1 , x2 , x3 } is a partial homogeneous system of parameters in F[V ] and hence also in F[V ]H . Assume by way of contradiction that F[V ]H is CohenMacaulay. Then x1 , x2 , x3 is a regular sequence in F[V ]H and therefore, since x3 u12 ≡ 0 (mod x1 , x2 ), we have u12 = f1 x1 + f2 x2

8 Polynomial Invariant Rings

145

for some f1 , f2 in F[V ]H . Clearly we may assume that the fi ’s are homogeneous of degree 2p − 2. As a K-invariant of this degree, we must have f1 = h0 + h1 N1 (y1 ) + h2 N2 (y2 ) + h3 N3 (y3 ), where hj ∈ F[x1 , x2 , x3 , x] for 0 ≤ j ≤ 3. Further, as a H-invariant, we have 0 = Δ(f1 ) = h1 Δ(N1 (y1 )) + h2 Δ(N2 (y2 )) + h3 N3 (y2 )) . Rewriting this expression we have (h1 + h2 + h3 )xp−1 = h1 x1p−1 + h2 x2p−1 + h3 x3p−1 . Expanding the right hand side as a linear combination of monomials we see that any non-zero term is a scalar multiple of a monomial from one of the hi ’s times the monomial xip−1 . But any such term must be divisible by xp−1 . Therefore, the degree of hi is at least p − 1, and so h1 = h2 = h3 = 0, from which we see f1 ∈ F[x1 , x2 , x3 , x]. The same argument applies to f2 . But this would mean u12 ∈ F[x1 , x2 , x3 , x], a contradiction. Lemma 8.0.10. Let S and T be subgroups of a group K. Then the cardinality of the set ST := {st | s ∈ S, t ∈ T } is given by |ST | =

|S||T | . |S ∩ T |

Proof. Let S := {si | 1 ≤ i ≤ [S : S ∩ T ]} be a set of left coset representatives for S ∩ T in S and let T := {tj | 1 ≤ j ≤ [T : S ∩ T ]} be a set of left coset representatives for S ∩ T in T . Define the set L := {si xtj | si ∈ S, x ∈ S ∩ T, tj ∈ T }. We see ST = L since clearly ST ⊆ L and L ⊆ ST . Suppose si xtj = si x tj  where si , si ∈ S, tj , tj  ∈ T and x, x ∈ S ∩ T . Therefore, −1 −1 −1  s−1 i si x = x tj  tj ∈ S ∩ T . Therefore, si si , tj  tj ∈ S ∩ T and thus si = si  and tj = tj  . Hence x = x . This shows that every element of ST can be written uniquely in the form si xtj with si ∈ S, tj ∈ T and x ∈ S ∩ T . Hence |S| |T | |S||T | · |S ∩ T | · |S∩T = |S∩T .   |ST | = [S : S ∩ T ] · |S ∩ T | · [T : S ∩ T ] = |S∩T | | | The following Theorem is due to Yinglin Wu [114]. Theorem 8.0.11. Let P be a p-subgroup of GL(V ). Suppose F[V ]P is a polynomial ring on the norms of the elements of some basis B = {y1 , y2 , . . . , yn } of V ∗ : P P F[V ]P = F[NP Py (y1 ), NPy (y2 ), . . . , NPyn (yn )] . 1

2

Then there exists an ordering of B with respect to which P is a Nakajima group. Proof. We begin by showing that |Py1 | · |Py2 | · · · |Pyr | = |W1 | · |W2 | · · · |Wr−1 | · |Py1 ∩ Py2 ∩ · · · ∩ Pyr | for all r = 2, 3, . . . , n where Wi := (Py1 ∩Py2 ∩· · ·∩Pyi )Pyi+1 for 1 ≤ i ≤ n−1.

146

8 Polynomial Invariant Rings

For r = 2, we need to show that |Py1 | · |Py2 | = |Py1 Py2 | · |Py1 ∩ Py2 |. This is immediate from Lemma 8.0.10. Now we proceed by induction on r. Hence we assume that |Py1 | · |Py2 | · · · |Pyr | = |W1 | · |W2 | · · · |Wr−1 | · |Py1 ∩ Py2 ∩ · · · ∩ Pyr | . Multiplying both sides of this equation by |Pyr+1 | gives |Py1 |·|Py2 | · · · |Pyr |·|Pyr+1 | = |W1 |·|W2 | · · · |Wr−1 |·|Py1 ∩Py2 ∩· · ·∩Pyr |·|Pyr+1 | . Applying Lemma 8.0.10 with S = Py1 ∩ Py2 ∩ · · · ∩ Pyr and T = Pyr+1 we get |Py1 ∩Py2 ∩· · ·∩Pyr |·|Pyr+1 | = |(Py1 ∩Py2 ∩· · ·∩Pyr )Pyr+1 |·|Py1 ∩Py2 ∩· · ·∩Pyr+1 | and therefore, |Py1 | · |Py2 | · · · |Pyr+1 | = |W1 | · |W2 | · · · |Wr | · |Py1 ∩ Py2 ∩ · · · ∩ Pyr+1 | , as required. In particular, since Py1 ∩ Py2 ∩ · · · ∩ Pyn = {e} we see that |Py1 | · |Py2 | · · · |Pyn | = |W1 | · |W2 | · · · |Wn−1 | . P P The hypothesis F[V ]P = F[NP Py1 (y1 ), NPy2 (y2 ), . . . , NPyn (yn )] implies that P i=1 deg(NPyi (yi )) = |P | by Corollary 3.1.6. Since deg(N(yk )) = |P |/|Pyi | for   all i, we have that ni=1 |P |/|Pyi | = |P | and therefore, ni=1 |Pyi | = |P |n−1 . n−1 Thus i=1 |Wi | = |P |n−1 . Since each of the sets Wi is a subset of P , we must have Wi = P for all i = 1, 2, . . . , n − 1. In particular, (Py1 ∩ Py2 ∩ · · · ∩ Pyn−1 )Pyn = P , i.e., Pn Pyn = P and thus |Pn Pyn | = |P |. But, again by |P |·|P | Lemma 8.0.10, |Pn Pyn | = |Pnn ∩Pyyn| = |Pn | · |Pyn | and hence |Pn | · |Pyn | = |P |. n Since B is unordered we similarly obtain |Pk |·|Pyk | = |P | for all k = 1, 2 . . . , n. Therefore,

n

n  k=1

|Pk | =

n  |P |n |P |n |P | = n = = |P | . |Pyk | |P |n−1 k=1 |Pyk |

k=1

Applying Lemma 8.0.6 shows that |P | = |Pn Pn−1 · · · P1 | and therefore, P = Pn Pn−1 · · · P1 . It only remains to show that there is some ordering of B with respect to which the group P is upper triangular. ∗ for all σ ∈ P } Define V0∗ := {0} and Vk∗ := {v ∈ V ∗ | (σ − 1)v ∈ Vk−1 ∗ ∗ for all k ≥ 1. Let s be minimal such that Vs = V . Choose an ordered basis {x1 , x2 , . . . , xn } of V ∗ which is compatible with the flag {0} = V0∗ ⊂ V1∗ ⊂ · · · ⊂ Vs∗ = V ∗ . Thus for each k with 1 ≤ k ≤ s, there exits a j such that {x1 , x2 , . . . , xj } is a basis of Vk∗ . Thus P is upper triangular with respect to the basis {x1 , x 2 , . . . , xn }. n Write xi = j=1 αij yj for i = 1, 2, . . . , n. For 1 ≤ k ≤ s, define

8.1 Stong’s Example

147

Wk := span{yj ∈ B | αij = 0 for some i with xi ∈ Vk∗ } . ∗ ∗ We n claim that Wk = Vk for all k = 1, 2, . . . , s. Take xi ∈ Vk . Then xi = j=1 αij yj with yj ∈ Wk for all j with αij = 0. Thus xi ∈ Wk . Therefore Vk∗ ⊆ Wk . For the opposite inclusion, let yj ∈ Wk . Then there exists some t such that ∗ xt ∈ Vk∗ and αtj = 0. Take any σ ∈ P . We want to show that (σ −1)yj ∈ Vk−1 . Write σ = σn σn−1 · · · σ1 with σa ∈ Pa for a = 1, 2, . . . , n. Since Pa normalizes  Pb when a < b, we may write σ = σk σn σn · · · σj+1 σj−1 · · · σ1 where σa ∈ Pa for    a = j + 1, j + 2, . . . , n. Thus (σ − 1)yj = σj σn σn · · · σj+1 σj−1 · · · σ1 (yj ) − yj = n (σj − 1)yj . But (σj − 1)xt = (σj − 1)( i=1 αti yi ) = αtj (σj − 1)yj . Since (σj − 1)xt ∈ Vk−1 and since αtj = 0, this shows that (σ − 1)yj = (σj − 1)yj ∈ Vk−1 . Therefore, yj ∈ Vk . Thus, Wk ⊆ Vk∗ . This shows that we may order the yj in such a way that B is compatible with the flag {0} = V0∗ ⊂ V1∗ ⊂ · · · ⊂ Vs∗ = V ∗ . Using this ordered basis for B we see that P is upper triangular. Hence P is a Nakajima group with respect to this ordered basis.  

In 1980, H. Nakajima [83] working over the prime field proved that, if Fp [V ]P is polynomial, then P is a Nakajima group. In the next section, we give an example, due to Stong, which shows that Nakajima’s result cannot be extended naively to other fields.

8.1 Stong’s Example Here we give an example due to R Stong that shows that Nakajima’s Theorem does not extend to larger fields. We work over the field Fq with q = p3 . We may suppose the field Fq has basis over Fp consisting of {1, ω, μ}. Let H be the group generated by the matrices ⎛ ⎞ ⎛ ⎞ 110 101 ⎝0 1 0⎠ and ⎝0 1 0⎠ , 001 001 and let G be the group generated by H and the matrix ⎛ ⎞ 1ωμ σ = ⎝0 1 0 ⎠ , 00 1 with respect to the basis {x, y, z} of V ∗ . We note that both groups are generated by reflections, but that G is not a Nakajima group, since we cannot choose a basis of V ∗ with respect to which each generating reflection is concentrated in a single column. It is not hard to see that Fq [V ]H = Fq [x, N (y), N (z)], where N (y) = p y −xp−1 y and N (z) = z p −xp−1 z. We calculate σ(N (y)) = N (y)−(ω p −ω)xp ,

148

8 Polynomial Invariant Rings

and σ(N (z)) = N (z)−(μp −μ)xp . From here we can construct two G invariants f1 = (μp −μ)N (y)−(ω p −ω)N (z) and f2 = N (y)p −(ω p −ω)(p−1) N (y)xp(p−1) . Using Lemma 2.6.3 it is not hard to see that {x, f1 , f2 } form a homogeneous system of parameters, and thus by Corollary 3.1.6 that Fq [V ] = Fq [x, f1 , f2 ] is a polynomial ring. We may also use Theorem 3.9.2 to see immediately that Fq [V ]G must be a polynomial ring.

8.2 A Counterexample For non-modular groups, we have the characterization of Shephard and Todd which asserts that F[V ]G is a polynomial ring if and only if the action of G on V is generated by reflections. It is known that this characterization fails for modular representations. For example, the representation described in Example 11.0.3 is generated by reflections but its ring of invariants is not polynomial. One of the most important open questions in modular invariant theory is to give a geometric characterization of modular representations of finite groups having a polynomial ring of invariants. In 1982, V. Kac [57] made the following conjecture. Conjecture 8.2.1. Let K be an algebraically closed field. Then K[V ]G is a polynomial ring if and only if each isotropy group Gv for v ∈ V is generated by reflections. In this section, we describe an unpublished counter-example to Kac’s conjecture due to Campbell, Hughes and Shank in 1995. We note that Example 11.0.3 is also an example of such a reflection group whose ring of invariants is not polynomial. In their classification of irreducible reflection groups, Kemper and Malle [63, Example 2.2] also gave a counter-example. Let F be an algebraically closed field of characteristic p and consider ⎧⎛ ⎫ ⎞ 1000 ⎪ ⎪ ⎪ ⎪ ⎨⎜ ⎬ ⎟ 0 1 0 0 ⎟ | a, b, c ∈ Fp . G= ⎜ ⎝a c 1 0⎠ ⎪ ⎪ ⎪ ⎪ ⎩ ⎭ cb01 G is a subgroup of GL4 (F) isomorphic to Cp3 . Note that G is a reflection group since it is generated by the following three reflections: ⎛ ⎛ ⎛ ⎞ ⎞ ⎞ 1000 1000 1000 ⎜0 1 0 0⎟ ⎜0 1 0 0⎟ ⎜0 1 0 0⎟ −1 −1 ⎜ ⎜ ⎟ ⎟ ⎟ α−1 := ⎜ ⎝1 0 1 0⎠ , β := ⎝0 0 1 0⎠ , γ := ⎝1 1 1 0⎠ . 0001 0101 1101 Theorem 8.2.2. For all v ∈ V , the isotropy group Gv is a reflection group.

8.3 Irreducible Modular Reflection Groups

149

Proof. Let v = (a1 , a2 , b1 , b2 ) ∈ V . It is clear that if a1 = a2 = 0, then Gv = G. If a1 = 0 and a2 = 0, then Gv is generated by α. If a1 = 0 and a2 = 0 then Gv is generated ⎞ if a1 = 0 and a2 = 0, then Gv is generated by ⎛ by β. Finally, 1 0 00 ⎜ 0 1 0 0⎟ ⎟ the reflection ⎜ ⎝ aa2 −1 1 0⎠ . 1 −1 aa12 0 1 Each of these subgroups is a reflection group and thus all of the isotropy subgroups are reflection groups.   Remark 8.2.3. Since each proper non-trivial isotropy subgroup is generated by a single transvection, the ring of invariants for each of these groups is a polynomial algebra. Theorem 8.2.4. F[V ]G is not a polynomial algebra. Proof. Let {x1 , x2 , y1 , y2 } denote the basis of V ∗ dual to the canonical basis of V . Thus α(y1 ) = y1 + x1 and β(y2 ) = y2 + x2 . If F[V ]G were a polynomial algebra, it would have four generators and the product of the degrees of the generators would be |G| = p3 . Observe that x1 and x2 are invariant and that G dimK F[V ]G 1 = 2. Thus if F[V ] is a polynomial algebra, it has one generator in degree p and one in degree p2 . We will show that F[V ]G does not have a generator in degree p. Consider the subgroup H of G generated by α and β. Then H is a Nakajima group isomorphic to Cp × Cp and F[V ]H ∼ = F[x1 , x2 , N1 , N2 ] where N1 = y1p − x1p−1 y1 and N2 = y2p − x2p−1 y2 . Since F[V ]G ⊆ F[V ]H , the degree p elements of F[V ]G are contained in the vector space spanned by N1 , N2 and the monomials in x1 and x2 of degree p. Since γ(N1 ) = N1 + (xp2 − x1p−1 x2 ) and γ(N2 ) = N2 +(xp1 −x2p−1 x1 ) we see that γ does not fix any non-zero linear combination of N2 and N1 . Hence F[V ]G does not have a generator in degree p and is not a polynomial algebra.   Remark 8.2.5. We show in Example 10.0.11 that F[V ]G is a hypersurface ring with two generators x1 and x2 of degree 1, a generator x1 N1 + x2 N2 of deG 2 gree p + 1 and two generators NG H (N1 ) and NH (N2 ) of degree p . Note that Proposition 11.0.1 can also be used to show this.

8.3 Irreducible Modular Reflection Groups In this section, we recall the work of Kemper and Malle [63]. Suppose G is represented on V over a field F of characteristic p dividing the order of G and that V is irreducible as a G-module. Of course, in modular representation theory, the irreducible representations are essentially few and far between, as we have just seen for Cp . Nevertheless, such representations have been classified. Using the classification, Kemper and Malle have proved the following

150

8 Polynomial Invariant Rings

Theorem 8.3.1. Suppose V is an irreducible representation of the modular group G. Then F[V ]G is a polynomial ring if and only if G is generated by reflections and if W is any non-trivial subspace of V , then F[V ]GW has a polynomial ring of invariants. We note that F[V ]GW polynomial implies that GW is generated by reflections. Remark 8.3.2. As observed by Kemper and Malle, the theorem is equivalent to the following. Suppose v is any point of V = F ⊗F V , where F denotes the algebraic closure of F, and that Gv denotes the stabilizer of v. Then F[V ]G is polynomial if and only if G is generated by reflections and F[V ]Gv is polynomial. Remark 8.3.3. Kemper and Malle, see [68], also proved that the invariant field, F(V )G , of an irreducible reflection group is purely transcendental. It is not sufficient to assume that each Gv is a reflection group. Example 8.3.4. Let G = Σ6 act on its 4-dimensional irreducible representation V in characteristic 2. Then Gv is generated by reflections for all non-zero v ∈ V , but F[V ]G is not polynomial. 8.3.1 Reflection Groups As noted above, a reflection σ ∈ GL(V ) is said to be a transvection if it is not diagonalizable. If it is diagonizable, it is said to be a homology. We denote by RLn (F) the subgroup of GLn (F) generated by its reflections. Similarly, we let RUn (F) denote the subgroup of GUn (F), the unitary group of dimension n over F, generated by its reflections. Then RLn (F) contains SLn (F) as a subgroup of index 2. Similarly respectively RUn (F) contains SUn (F) as a subgroup of index 2. We denote by GOn (F) the general orthogonal group over F and by SOn (F), the special orthogonal group over F. And we denote by Ωn± (F) the commutator subgroup of GOn (F). Finally, we denote by Gi the i-th group in the list of finite reflection groups given by Shephard and Todd [101] and by Wp (Gi ) the mod p reduction of its reflection representation. The irreducible subgroups of GLn (F) generated by reflections were classified by Wagner [109] and by Zalesski˘ı, and Sereˇzkin [115], for n ≥ 3. Theorem 8.3.5. Let V be a vector space of dimension n ≥ 3 over F of characteristic p > 2 and let G denote a subgroup of GL(F) generated by reflections such that V is irreducible and primitive as a G-module. Then one of the following holds: 1. G = RLn (F), or G = RUn (F); ± 2. Ωn± (F)  G ⊆ GO± n (F), G = SOn (F); 3. G = Σn+1 for p  (n + 1);

8.3 Irreducible Modular Reflection Groups

151

4. G = Σn+2 for n ≥ 5 and p  (n + 2); 5. G = Wp (Gi ) for 23 ≤ i ≤ 37, i = 25, 26, 32 and p  |G| or p ∈ {3, 5, 7} corresponds to the columns marked p or np in [69, Table 6.3]; 6. p = 5, n = 3 and G is the group 3 · A7 × 2 (sometimes denoted EJ3 (5)); 7. p = 3, n = 4 and G is the group 4 · L3 (4) : 22 (sometimes denoted J4 (4)). 8.3.2 Groups Generated by Homologies of Order Greater than 2 These groups and representations were classified by Mitchell in characteristic 0 and by Wagner in positive characteristic, see [108]. Theorem 8.3.6. Let V be a vector space of dimension n ≥ 3 over F of characteristic p > 2 and let G denote a subgroup of GL(F) generated by pseudoreflections, containing homologies of order greater than 2 such that V is irreducible and primitive as a G-module. Then one of the following holds: 1. SLn (F) ⊂ G ⊂ GLn (F) or SUn (F ⊂ G ⊂ GUn (F), G = RLn (F), RUn (F); 2. G = Wp (G25 ) and p = 3; 3. G = Wp (G26 ) and p = 2, 3; 4. G = Wp (G32 ) and p = 3. 8.3.3 Groups Generated by Transvections There is no analogue for these groups in characteristic 0. We refer to [58]. Theorem 8.3.7. Let V be a vector space of dimension n over F, a finite field of characteristic p. Let G denote a subgroup of GL(F) generated by transvections such that V is irreducible and primitive as a G-module. Then one of the following holds: 1. G = SLn (F), Spn (F), SUn (F) and p = 2, n = 3 ; 2. G = SOn (F), p = 2 and n ≥ 4 is even; 3. G = Σn+1 or Σn+2 , p = 2 and n ≥ 6 is even; 4. G = Σn+2 for n ≥ 5 and p  (n + 2); 5. G = SL2 (F), p = 5 and n = 3; 6. p = 2, n = 3 and G is the group 3 · A6 ; 7. p = 2, n = 6 and G is the group 3 · U4 (3) : 22 .

9 The Transfer

In this chapter, we consider in detail the transfer (also called the trace) map introduced in §1.2. Let H be a subgroup of the finite group G. Choose a set of left coset representatives for H in G. We denote this set of representatives by G/H. Thus G = σ∈G/H σH is a decomposition of G into left cosets. There is an extensive theory considering the relative versions of the results of this chapter, see Fleischmann [38] or Fleischmann and Shank [41]. As in §1.2, we define the relative transfer H −→  K[V ]G TrG H : K[V ] f −→ σ(f ) . σ∈G/H G H and take an arTo see that the image of TrG H lies in K[V ] , let f ∈ K[V ] bitrary group element σ0 ∈ G. Write G/H = {σ1 , σ2 , . . . , σr } where r is the index of H in G. Then {σ0 σ1 , σ0 σ2 , . . . , σ0 σr } is another set of left coset representatives for H in G. Therefore, there exist τ1 , τ2 , . . . , τr ∈ H such that {σ0 σ1 , σ0 σ2 , . . . , σ0 σr } = {σ1 τ1 , σ Therefore, σ0 (TrG 2 τ2 , . . . , σr τr }.  H (f )) = r r r r σ0 ( i=1 σi (f )) = i=1 σ0 σi (f ) = i=1 σi τi (f ) = i=1 σi (f ) = TrG H (f ).

Lemma 9.0.1. The map TrG H is independent of the choice of coset representatives.   The most important of these maps is the map TrG {e} which we denote by G Tr . In general, the transfer map does not behave well with respect to products. However, if f1 ∈ F[V ]G and f2 ∈ F[V ]H then we have   TrG σ(f1 f2 ) = σ(f1 )σ(f2 ) H (f1 f2 ) = σ∈G/H

=



σ∈G/H

σ∈G/H

f1 σ(f2 ) = f1



σ(f2 ) = f1 TrG H (f2 ).

σ∈G/H

H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, 153 Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 9, © Springer-Verlag Berlin Heidelberg 2011

154

9 The Transfer

G Thus TrG H is an F[V ] -module homomorphism. Also, note that the transfer map preserves degree.  |G| Notice that if f ∈ F[V ]G , then TrG H (f ) = σ∈G/H σ(f ) = |H| f = [G : H]f . Thus the composition TrG

H F[V ]G → F[V ]H −→ F[V ]G

is just multiplication by [G : H]. Therefore, if [G : H] is invertible in F, then 1 TrG : K[V ]H → K[V ]G [G : H] H is a projection operator. This projection onto the invariants is known as the Reynolds Operator. In particular, using the Reynolds Operator, we see that for non-modular representations, we have F[V ]H = F[V ]G ⊕ (ker TrG H ). For this reason, the transfer is well understood and accordingly less interesting in the non-modular case. The following well-known lemma is very useful in studying the image of the transfer for modular groups. Lemma 9.0.2. Let q = pr be a prime power and suppose that  is a positive integer. Then   −1, if q − 1 divides ; c = 0, if q − 1 does not divide . c∈Fq

Proof. It is sufficient to sum over the non-zero elements of Fq . Let η be a generator for the group of units of Fq . Hence Fq \ {0} = {η 0 , η, η 2 , . . . , η q−2 } and, therefore, q−2 q−2     i  c = (η ) = (η )i . c∈Fq

i=0

i=0

Every non-zero element of Fq , including in particular η , is a root of the polynomial xq−1 −1 = (x−1)(1+x+. . .+xq−2 ). If q −1 divides , then η  = 1 and the sum is q − 1. If q − 1 does not divide , then η  = 1 and therefore, η  is a root of 1 + x + . . . + xq−2 . Hence if q − 1 does not divide , then the sum is zero.   Remark 9.0.3. In the above lemma,  is a positive integer. However, it will be convenient to consider the case where  = 0. In that case, the sum, c∈Fq c0 , involves 00 and so is not well-defined. In most of the applications we will use, the sum will most naturally be 0 and so we will consider 00 to be equal to 1. Thus even though q − 1 does divide  when  = 0, the sum in our applications will be 0.

9 The Transfer

155

Example 9.0.4. We consider the two dimensional irreducible representation, V2 , of the cyclic group of order p over a field F of characteristic p. We choose a basis {x, y} for V2∗ and a generator σ of Cp such that σ(y) = y + x and σ(x) = x. Since the ring of invariants F[V2 ]Cp = F[x, N = y p −xp−1 y] is a polynomial subring, we have that F[V2 ] = F[x, y] is a free F[V2 ]Cp -module. In fact, it is Cp j y . We use this Hironaka decomposition easy to see that F[V2 ] = ⊕p−1 j=0 F[V2 ] to study the transfer map: Cp j TrCp : F[V2 ] = ⊕p−1 y → F[V2 ]Cp . j=0 F[V2 ]

p−1 p−1 We have TrCp = i=0 σ i and therefore, TrCp (f (x, y)) = i=0 f (x, y + ix). It is easy to show that σi (y) = y + ix. Thus σi (y j ) = (y + ix)j and therefore TrCp (y j ) =

p−1 

(y + ix)j

i=0

=

j  p−1   i=0 t=0

 j t−j y (ix)t t

j   p−1  j t−j t  t = i). y x( t t=0 i=0

Thus by Lemma 9.0.2, we see that TrCp (y j ) = 0 unless j = p − 1 in which p−1 p−1 Cp p−1 case we have Tr (y ) = p−1 y 0 xp−1 ( i=0 ip−1 ) = −xp−1 . p−1 Given f ∈ F[V2 ], we write f = j=0 fj y j where fj ∈ F[V2 ]Cp for each p−1 p−1 j = 0, 1, . . . , p − 1. Then TrCp (f ) = j=0 TrCp (fj y j ) = j=0 fj TrCp (y j ) = −fp−1 xp−1 . Thus the image of TrCp is the principal ideal of F[V2 ]Cp generated by xp−1 . Example 9.0.5. Take G to be the symmetric group on 3 letters and K to be any field. Let G act on V = K3 by permuting the basis {e1 , e2 , e3 } of V . Write {x, y, z} for the dual basis of V ∗ . Then G also permutes {x, y, z}. Then K[V ]G = K[s1 , s2 , s3 ] where s1 = x + y + z, s2 = xy + xz + yz and s3 = xyz. We have G = {e, (12), (13), (23), (123), (132)} where for example (123) · ae1 + be2 + ce3 = ae2 + be3 + ce1 and (123) · y = z. Then TrG (f ) = f + (12) · f + (13) · f + (23) · f + (123) · f + (132) · f . Thus G Tr (f (x, y, z)) = f (x, y, x) + f (y, x, z) + f (z, y, x) + f (x, z, y) + f (y, z, x) + f (z, x, y). Let N denote the normal subgroup of index 2 in G, the alternating group on 3 letters. Thus TrN (f ) = f + (123) · f + (132) · f = f (x, y, z) + f (y, z, x) + f (z, x, y). The coset decomposition of G with respect to N is G = N  (12)N = {e, (123), (132)}  {(12), (13), (23)}. We have TrG N (f ) = f + (12) · f for f ∈ F[V ]N . Thus TrG N (f (x, y, z)) = f (x, y, z) + f (y, x, z).

156

9 The Transfer

Now consider the subgroup H = {e, (12)} of G of order 2 which fixes e3 and z. Then TrH (f ) = f + (12) · f = f (x, y, z) + f (y, x, z). Decomposing G as a union of left H-cosets gives G = H  (123)H  (132)H = {e, (12)}  {(123), (13)}  {(132), (23)}. Hence TrG H (f ) = f + (123) · f + (132) · f for f ∈ F[V ]H . Therefore, TrG H (f (x, y, z)) = f (x, y, z) + f (y, z, x) + f (z, x, y). To study the full transfer homomorphism, TrG , we exploit the block basis and corresponding Hironaka decomposition given in Proposition 6.1.1: K[V ] ∼ = K[V ]G ⊕ K[V ]G x ⊕ K[V ]G y ⊕ K[V ]G xy ⊕ K[V ]G x2 ⊕ K[V ]G x2 y . Let f ∈ K[V ] and write f = f0 + f1 x + f2 y + f3 xy + f4 x2 + f5 x2 y where fi ∈ K[V ]G for all i = 0, 1, . . . , 5. It is easy to compute that TrG (1) = |G| = 6, TrG (x) = TrG (y) = 2(x + y + z) = 2s1 , TrG (xy) = 2(xy + xz + yz) = 2s2 , TrG (x2 ) = 2(x2 + y 2 + z 2 ) = 2(s21 − 2s2 ), and TrG (x2 y) = (x2 y + x2 z + xy 2 + xz 2 + y 2 z + yz 2 ) = s1 s2 − 3s3 . Thus TrG (f ) = TrG (f0 + f1 x + f2 y + f3 xy + f4 x2 + f5 x2 y) = TrG (f0 ) + TrG (f1 x) + TrG (f2 y) + TrG (f3 xy) + TrG (f4 x2 ) + TrG (f5 x2 y) = f0 TrG (1) + f1 TrG (x) + f2 TrG (y) + f3 TrG (xy) + f4 TrG (x2 ) +f5 TrG (x2 y) = 6f0 + 2f1 s1 + 2f2 s1 + 2f3 s2 + 2f4 (s21 − 2s2 ) +f5 (s1 s2 − 3s3 ) . Lemma 9.0.6. For G a finite group, if N ≤ H ≤ G is a sequence of subgroups G H of G, then TrG N = TrH ◦ TrN . We may use right cosets to define another useful map. If H is a subgroup of G, choose a set Ω of right coset representatives for H in G. and define  G G Tr H : K[V ] → K[V ] by TrH (f ) = σ∈Ω σ(f ). This map depends upon the choice of coset representatives used. However, for all possible choices we have the following factorization of TrG N. Lemma 9.0.7. Let N be a normal subgroup of the finite group G and suppose H G H is another subgroup of G with N ≤ H ≤ G. Then TrG N = TrN ◦TrH .

9 The Transfer

157

Proof. Write Ω = σ1 , σ2 , . . . , σr for the right coset representatives of H in G G r chosen to define the map Tr H . Then we have the decomposition G = i=1 Hσi of G into right H-cosets. Let H = sj=1 τj N be a decomposition of H into left N -cosets. Then G = ri=1 sj=1 τj N σi . Since N is normal in G, this gives G = ri=1 sj=1 τj σi N . Thus for any f ∈ K[V ]N , TrG N (f ) =

s r  

τj σi (f ) =

i=1 j=1

We claim that

r s  

τj σi (f ) =

j=1 i=1

s 

G

(f ) . τj Tr H

j=1

s

G H G

j=1 τj TrH (f ) = TrN (TrH (f )). In order to see this, we only G

G (f ) is N H (f ) lies in the domain of TrG , i.e., that Tr Tr H N

need show that invariant. To see this, let n ∈ N be arbitrary and note that f is N -invariant.

G (f ) = n · r σi (f ) = Thus writing ni = σi−1 nσi ∈ N we have n · Tr H i=1 G r r r

(nσ )(f ) = σ n (f ) = σ (f ) = Tr (f ), as required.   i i i i H i=1 i=1 i=1 The following corollary is immediate. H N N Corollary 9.0.8. TrG N (K[V ] ) ⊆ TrN (K[V ] ).

Example 9.0.9. We continue with Example 9.0.5. Examples of factorizations N G H G of TrG include TrG (f ) = TrG N (Tr (f )) = TrH (Tr (f )) and Tr (f ) = G

H (f )) for f ∈ K[V ]. This latter factorization gives TrG (f (x, y, z)) = TrH (Tr H Tr (h(x, y, z)) = h(x, y, z) + h(y, x, z) where for one choice of coset represen G (f (x, y, z)) = f (x, y, z)+f (y, z, x)+f (z, y, x). tatives, we have h(x, y, z) = Tr H

G As we noted above, the transfer map TrG H is a homogeneous F[V ] -module G H homomorphism. In particular, its image, TrH (F[V ] ), is a graded F[V ]G H submodule of F[V ]G , that is to say, the image of the transfer, TrG H (F[V ] ), is a G homogeneous ideal of F[V ] . We will be interested in describing the algebraic −1 (X) ⊂ V ) corresponding to subset, X, of V //G (more precisely, its lift πV,G this ideal. We will assume throughout the rest of this chapter that the base field has positive characteristic p unless stated otherwise.

Theorem 9.0.10. Let N be a normal subgroup of the finite group G and suppose F has characteristic p. Then  σ N V V V (TrG N (F[V ] )) = order(σN )=p

where the union is over all quotient group elements σN ∈ G/N of order p.

158

9 The Transfer

Proof. First, we prove σ

N V(TrG N (F[V ] )) ⊇ ∪order(σN )=p V . σ

Let v ∈ ∪order(σN )=p V , i.e., suppose there exists σ ∈ G \ N , with σ p ∈ N and σv = v. Let H denote the subgroup of G generated by N and σ. Thus [H : N ] = p. Take f ∈ F[V ]N . Then p−1  (f ) = σ i (f ) TrH N i=0

and (TrH N (f ))(v) =

p−1  i=0

 p−1

f (σ −i v) =

f (v) = pf (v) = 0.

i=0

By the previous corollary, (TrG N (f ))(v) = 0 also. Therefore,   N v ∈ V V TrG N (F[V ] ) . σ

N Next, we prove the opposite inclusion, V(TrG N (F[V ] )) ⊆ ∪order(σN )=p V . σ Take v ∈ V \ ∪order(σN )=p V . By Corollary 2.1.3, there exists h ∈ F[V ] such that 1 if w ∈ N v, h(w) = 0 if w ∈ Gv \ N v.

Notice that it can happen that σ ∈ / N but σv ∈ N v, i.e., σv = τ v for some τ ∈ N and some σ ∈ G \ N . In this case, τ −1 σ ∈ Gv and therefore, σ ∈ N · Gv where since N is normal in G, the set N · Gv is a subgroup of G. Conversely, if σ is any element of N · Gv , then there exist τ ∈ N and σ1 ∈ Gv such that σ = τ σ1 and thus σv = τ v ∈ N v. In conclusion, for σ ∈ G, we have σv ∈ N v if and

only if σ ∈ N · Gv . Let f := τ ∈N τ (h) ∈ F[V ]N . For all τ ∈ N , we have τ −1 w ∈ N v if and only if w ∈ N v and thus  1 if w ∈ N v, −1 f (w) = h(τ w) = 0 if w ∈ Gv \ N v. τ ∈N Let G = ti=1 σi N be a decomposition of G into left N -cosets. Then (TrG N (f ))(v) =

t 

f (σi−1 v) = |{i | σi−1 v ∈ N v}|

i=1

= |{i | σi−1 ∈ N · Gv }| = |{i | σi ∈ N · Gv }| . Consider a right coset N σ of N and suppose this right coset meets N · Gv . Then we have τ1 σ = τ σ1 where τ1 , τ ∈ N and σ1 ∈ Gv . Now let τ2 σ be any

9 The Transfer

159

other element of the coset N σ. Then τ2 σ = τ2 τ1−1 τ σ1 is also in N · Gv . Thus if the coset N σ meets N · Gv , it is contained in N · Gv . Since N is normal, every right coset is also a left coset: N σ = σN , and thus we see that  N · Gv = σi N . σi ∈N ·Gv

Therefore, (TrG N (f ))(v) = |{i | σi ∈ N · Gv }| = |

N · Gv | N

Gv | by the second isomorphism theorem for groups N ∩ Gv Gv =| |. Nv =|

Thus it remains to prove that p does not divide the index of Nv in Gv . Assume by way of contradiction that p does divide [Gv : Nv ]. Then there v exists σ1 ∈ Gv such that σ1 Nv has order p in G Nv , i.e., σ1 ∈ Gv \ Nv and p / N . Thus σ1 N has order p σ1 ∈ Nv . Since Nv = Gv ∩ N , we see that σ1 ∈ σ1 as an element of G/N . Also, v ∈ V since σ1 ∈ Gv . This contradicts our Gv original assumption on v and this contradiction shows that p does divide | N |. v G G N / V V (TrN (F[V ] )).   Therefore (TrN (f ))(v) = 0 and hence v ∈ Of course, the most important case of the above Corollary is when N = {e}. In that case we have    σ V V TrG (F[V ]) = V . order(σ)=p

Thus the subvariety of V corresponding to the ideal of F[V ] generated by the elements in the image of the transfer consists of precisely those points x in V for which p divides the order of the isotropy group Gx . Example 9.0.11. We continue with Example 9.0.9. We have seen that the image of TrΣ3 is the ideal in K[V ]Σ3 = K[s1 , s2 , s3 ] generated by 6, 2s1 , 2s2 , 2(s21 − 2s2 ), s1 s2 − 3s3 . Thus if the characteristic of K is not 2 nor 3, then TrΣ3 is surjective. If K has characteristic 2, then the image of the transfer is the principal ideal generated by s1 s2 − s3 = (x + y + z)(xy + xz + yz) − xyz = x2 y + x2 z + xy 2 + xz 2 + y 2 z + yz 2 = x2 (y + z) + x(y + z)2 + yz(y + z) = (y + z)(x2 + xy + xz + yz) = (y + z)(x(x + y) + z(x + y)) = (y + z)(x + y)(x + z)

160

9 The Transfer

Thus the image of the transfer vanishes precisely on the union of planes: (12) (13) (23) V ∪V ∪V = {(a, a, c) ∈ V | a, c ∈ K} ∪ {(a, b, a) ∈ V | a, b ∈ K} ∪ {(a, b, b) ∈ V | a, b ∈ K}. Now we consider the situation when the characteristic of K is 3. In that case, the image of the transfer is generated by the two invariants s1 and s2 . Taking the two equations x + y + z = 0 and xy + xz + yz = 0 and substituting z = −(x + y) into the second gives 0 = xy − x(x + y) − y(x + y) = −(x2 + xy + y 2 ) = −(x2 − 2xy + y 2 ) = −(x − y)2 Thus if both s1 and s2 vanish at a point v = (a, b, c) ∈ V , we must have a = b and similarly, b = c. Therefore, we find that in characteristic 3, the ideal TrΣ3 (K[V ]) cuts out the line {(a, a, a) ∈ V | a ∈ K}. Of course, this line is precisely the set of points of V fixed by the subgroup N = {e, (123), (132)} of order p = 3. For any σ ∈ G, we denote by Pσ the ideal of F[V ] generated by the set (σ − 1)V ∗ := {σ · x − x | x ∈ V ∗ }. Lemma 9.0.12. Let σ ∈ G. Then σ

V V (Pσ ) = V . In fact,

σ

Pσ = IF[V ] (V ). σ

Proof. First, let v ∈ V and take x ∈ V ∗ . Then ((σ − 1) · x)(v) = (σ · x)(v) − x(v) = x(σ −1 v) − x(v) = x(v) − x(v) = 0. Thus v ∈ V V (Pσ ). For the opposite inclusion, take v ∈ V V (Pσ ) and consider a basis {x1 , x2 , . . . , xn } of V ∗ . Write v = (v1 , v2 , . . . , vn ) and w := σ −1 v = (w1 , w2 , . . . , wn ) in terms of (the dual of) this basis. Then 0 = ((σ − 1) · xi )(v) = (σ · xi )(v) − xi (v) = xi (σ −1 v) − xi (v) = xi (w) − xi (v) = wi − vi σ

for all i = 1, 2, . . . , n. Therefore v = w = σ −1 v and thus v ∈ V . σ Having shown V V (Pσ ) = V , the final statement in the lemma is equivalent to the assertion that Pσ is a radical ideal. Since Pσ is generated by homogeneous elements of degree 1, it is in fact a (homogeneous) prime ideal and thus is certainly radical.  

9 The Transfer

161

Since Pσ is a prime ideal, by the “lying over” Theorem 2.5.2 (1), Pσ ∩F[V ]G is also a prime ideal. Let f, h ∈ F[V ]. Then (σ − 1)(f h) = σ(f )σ(h) − f h = σ(f )σ(h) − σ(f )h + σ(f )h − f h = σ(f ) · (σ − 1)(h) + h · (σ − 1)(f ) ∈ Pσ . Hence (σ − 1)F[V ] ⊂ Pσ . Specializing of the result of Theorem 9.0.10 to the case N = {e}, we have the following equality of ideals:  Im TrG = (∩order(σ)=p Pσ ) ∩ F[V ]G . Proposition 9.0.13. Let G be a permutation group. Then Im TrG is a radical ideal. Proof. Recall that {OG (m) | m is a monomial of F[V ]} is a vector space basis for F[V ]G . Since for any monomial, m, TrG (m) = |Gm |OG (m), we see that the set {OG (m) | m is a monomial of F[V ], p does not divide |Gm |} is a vector space basis for Im TrG . Im TrG . Express the invariant Suppose that f ∈ Im TrG and write f r ∈  t f as a linear combination of orbit sums: f = i=1 ci OG (mi ) where ci ∈ F and mi is a monomial in F[V ] for all i = 1, 2, . . . , t. Choose m ∈ N large m m t m enough that pm ≥ r. Then f p = i=1 cpi OG (mpi ) ∈ Im TrG . Since V is a s permutation representation, it follows that tGm =mGm formall monomials m of F[V ] and for all positive integers s. Thus i=1 cpi OG (mpi ) ∈ Im TrG implies m  that cpi = 0 for all i for which p divides |Gmi |. Hence f = ti=1 ci OG (mi ) expresses f as a linear combination of elements of Im TrG and thus f ∈ Im TrG .   Suppose now that G = τ  is a cyclic permutation group with p dividing |G|. Define σ := τ |G|/p . Then the subgroup generated by σ contains all the elements of order p in G. Every element of order p in G is of the form σ r σ σr for some r with 1 ≤ r < p. Clearly, V ⊆ V . If we choose integers s ≥ 1 and t ≤ 0 such that rs + pt = 1, then (σr )s = σ(σ p )−t = σ and we see   p−1 σr σr σ σ that V ⊆ V . Therefore, V V TrG (F[V ]) = r=1 V = V . which is a  subspace of V and hence is an irreducible variety. Thus Im TrG = Im TrG is a prime ideal. We may give the same argument algebraically as follows. Consider such an element of G of order p which we again write as σ r where 1 ≤ r < p. From the factorization σ r − 1 = (1 + σ + σ 2 + · · · + σ r−1 )(σ − 1), we see that (σr − 1)V ∗ ⊆ (σ − 1)V ∗ and therefore, Pσr ⊆ Pσ . Again, writing (σ r )s = σ for an appropriate positive integer s shows that Pσ = P(σr )s ⊆ Pσr . Hence Pσ =  p−1 Pσr ) ∩ Pσr for all r = 1, 2, . . . , p − 1. Therefore, Im TrG = Im TrG = (∩r=1 G G F[V ] = Pσ ∩ F[V ] is a prime ideal. We record this result as Proposition 9.0.14. If G is a modular cyclic permutation group, then

162

9 The Transfer

Im TrG is a prime ideal. Example 9.0.15. Consider the regular representation Vp of Cp = σ defined over Fp . As usual, we choose a triangular basis {x1 , x2 , . . . , xp } for Vp∗ dual C to the basis {e1 , e2 , . . . , ep } of Vp . Then Vpσ = Vp p = spanFp {ep } is a one dimensional space. By definition, the ideal Pσ of F[Vp ] is generated by the degree 1 elements (σ − 1)x as x varies over Vp∗ . We have xi−1 if 2 ≤ i ≤ p; (σ − 1)xi = σ(xi ) − xi = 0 if i = 1. Thus Pσ = (x1 , x2 , . . . , xp−1 ) is the ideal generated by the linear functionals Cp which vanish on the fixed line V and Im TrCp = (x1 , x2 , . . . , xp−1 )∩F[Vp ]Cp . Next, we will show that for any modular representation of G, the ideal TrG (F[V ]) is a non-zero proper subset of the irrelevant ideal, F[V ]G + := G G ⊕∞ F[V ] , of F[V ] . d=1 d If f is invariant, then TrG (f ) = |G|f = 0 since p divides |G|. In particular, this means that in degree 0, TrG (F[V ]) contains only 0. This proves that TrG (F[V ]) is a subset of the irrelevant ideal. In fact, it is always a proper subset of the irrelevant ideal since by Theorem 9.0.10, it cuts out the variety of points v whose isotopy group Gv contains an element of order p. Thus V V (TrG (F[V ])) = {v ∈ V | p divides |Gv |}. σ Therefore, TrG (F[V ]) = F[V ]G = {0} for every element of + if and only if V order p ∈ G. But by Lemma 4.0.1, we see that the subspace V σ is never zero if the order of σ is p. Of course, since G is a modular group, there will exist elements in G of order p. The following lemma which is a consequence of Artin’s Theorem on Characters (see, for example, [75, VI Theorem 4.1]) will be used to show that TrG is never the zero map. Lemma 9.0.16. Let K be any field and let Γ := {φα | α ∈ A} be any set of (distinct) field automorphisms of K. Then Γ is a linearly independent subset of the vector space of all maps from K to K. Proof. Assume by way of contradiction that there is a linear relation c1 φα1 + c2 φα2 + · · · + ct φαt = 0 among the elements of Γ . Furthermore, we assume that t is minimal so that any subset of Γ containing t − 1 or fewer elements is linearly independent. Thus each of c1 , c2 , . . . , ct must be non-zero. Note that t ≥ 2 since the zero map is not a field automorphism. Let a ∈ K be arbitrary. Evaluating the above linear relation at x ∈ K and multiplying by φα1 (a) yields the equation

9 The Transfer

163

c1 φα1 (a)φα1 (x) + c2 φα1 (a)φα2 (x) + · · · + ct φα1 (a)φαt (x) = 0 which is valid for all x ∈ K. On the other hand, if we evaluate the linear relation at the point ax we get c1 φα1 (a)φα1 (x) + c2 φα2 (a)φα2 (x) + · · · + ct φαt (a)φαt (x) = 0 for all x ∈ K. Subtracting we find c2 (φα1 (a) − φα2 (a))φα2 (x) + c3 (φα1 (a) − φα3 (a))φα3 (x) + . . . + ct (φα1 (a) − φαt (a))φαt (x) = 0 which holds for all x ∈ K. By the minimality of t, we must have φα1 (a) − φαi (a) = 0 for all i = 2, 3, . . . , t. In particular, φα1 (a) = φα2 (a). But since a ∈ K was arbitrary, this implies φα1 = φα2 . This contradiction shows that Γ is linearly independent.   Corollary 9.0.17. Let V be a representation of the finite group G. The image of the transfer TrG : F[V ] → F[V ]G is non-zero. Proof. Each σ ∈ G defines an automorphism, φσ of the ring F[V ] given by φσ (f ) = σ(f ). We may extend these ring automorphisms to field automorphisms of F(V ), the quotient field of F[V ]. We do this  by defining φσ (f /h) := φσ (f )/φσ (h). Then, by the preceding lemma, σ∈G φσ = 0,  φ (f /h) =  0. Now define i.e., there exists f, h ∈ F[V ] such that σ∈G σ

h := σ∈G σ(h) and f  := f σ∈G\{e} σ(h). Then f  ∈ F[V ] and h ∈ F[V ]G     and f /h = f  /h . Therefore φσ (f  /h ) = σ∈G σ∈G (φσ (f )/φσ (h )) =  G G ( σ∈G φσ (f  ))/h = Tr (f  )/h is non-zero. Hence Tr (f  ) = 0.   σ

Note that by Theorem 9.0.10, this corollary shows that ∪order(σ)=p V = V , i.e., there must always be some points of V whose isotropy group is nonmodular. Proposition 9.0.18. Let N ≤ H ≤ G be a sequence of subgroups of G where N is normal in G and suppose that p does not divide the index of H in G. H G Then Im TrG N = Im TrN ∩F[V ] . H G Proof. First, suppose that f ∈ Im TrH N ∩F[V ] and write f = TrN (h) where G N G h ∈ F[V ] . Since f ∈ F[V ] , we have TrH (f ) = [G : H]f and thus G H G TrG N (h/[G : H]) = TrH (TrN (h/[G : H])) = TrH (f /[G : H]) = f .

For the opposite inclusion, we consider the factorization from Lemma 9.0.7: G

. Suppose that f ∈ Im TrG and write f = TrG (h) where = TrH ◦Tr

TrG N

N

H

G

N

H

h ∈ F[V ]N . Then f = TrH N (TrH (h)) ∈ Im TrN .

N

 

164

9 The Transfer

Corollary 9.0.19. Let P be a p-Sylow subgroup of G and let N ⊂ P be a G normal subgroup of G. Then the two ideals Im TrP N and Im TrN have the same height. Proof. Since F[V ]P is integral over F[V ]G , we may apply “going-up” and “going-down” (Theorem 2.5.2). Since the above lemma shows that Im TrP N lies over Im TrG   N , the result follows.

9.1 The Transfer for Nakajima Groups We want to describe the image of the transfer for a Nakajima group . We begin with some preliminaries. The following lemma is known as Lucas’ Lemma. Lemma 9.1.1. Let p be prime and let m = m0 + m1 p + m2 p2 + · · · + ms ps and i = i0 + i1 p +i2 p2 +· · · + is ps be the p-adic expansions of two non-negative integers m and i. Then        m1 ms m m0 ··· (mod p). ≡ i0 i1 is i Proof. Considering the expression (1 + y)m as a polynomial with coefficients in Fp we have m   s   j m j y = (1 + y)mj p j j=0 j=0

=

s 

j

(1 + y p )mj

j=0

=

 mj  s   mj

j=0

=

k

 p−1  s   mj

j=0

=

k=0

k=0

p−1 p−1   k0 =0 k1 =0

k

y kp

j



y kp

j



    p−1   m0 m1 ms k0 +k1 p+...ks ps ··· y ··· . k0 k1 ks ks =0

Comparing the coefficient of y i in the first and last expressions gives the result.   In fact, the following generalization is true. Lemma 9.1.2. Let q = pr be a prime power and let m = m0 +m1 q+. . .+ms q s be the q-adic expansion of m and ir = ir,0 + ir,1 q + . . . + ir,s q s is the q-adic expansion of ir . Then

9.1 The Transfer for Nakajima Groups



m i1 , . . . , ik



 ≡

m0 i1,0 , . . . , ik,0

  ·





m1 ms ··· i1,1 , . . . , ik,1 i1,s , . . . , ik,s

165

 (mod p).

Remark 9.1.3. The preceding lemma is essentially due to Dickson   —e see [28]. There he also proves that the highest power of p dividing m is p where e I is the total amount carried when the addition i1 + · · · + ik = m is performed base p. Definition 9.1.4. Let m be a non-negative integer. We denote by αp (m) the sum of the digits s in the p-adic expansion of m: αp (m) := m0 + m1 + · · · + ms where m = i=0 mi pi and 0 ≤ mi < p for all i = 0, 1, . . . , s. If W is an Fp subspace, i.e., if W is closed under Let W be a subset of V ∗ . addition, we define d(W ) := x∈W \{0} x. Now we prove a generalization of Lemma 9.0.2. Lemma 9.1.5. Let V be an F vector space and let W be a finite subset of V ∗ which is closed under addition, so that W is a vector space over Fp . Put m := dimFp (W ). Then in F[V ] we have  xs = 0 unless p − 1 divides s and αp (s) ≥ m(p − 1). 1. 

x∈W

2.

m

xp

−1

= d(W ).

x∈W

Proof. We use the convention that 00 = 1. Thus when s = 0, we have  0 m x∈W x = p · 1 = 0. We assume for the following that s ≥ 1. The proof is by induction on m. Let {x1 , x2 , . .  . , xm } be an Fp basis of W . For m = 0, the result is clear. For m = 1, we have x∈W xs = c∈Fp (cx1 )s =  xs1 c∈Fp cs . Now let η be a generator for the cyclic group of non-zero elements   p−2 j s of Fp . Then c∈Fp cs = c∈Fp \{0} cs = j=0 (η ) . If p − 1 divides s, then  sj s η = 1 for all j and c∈Fp c = p − 1 = −1. Since every non-zero element of Fp is a root of the polynomial xp −1 = (x−1)(xp−2 +xp−3 +· · ·+x+1), if p−1 does not divide s, then η s = 1 and thus η s is a root of (xp−2 +xp−3 +· · ·+x+1). Thus if p − 1 does not divide s, then c∈Fp cs = 0. Therefore, for m = 1 we have  −xs1 , if p − 1 divides s; s x = 0, if p − 1 does not divide s. x∈W Suppose m > 1 and let W  denote the Fp vector space spanned by {x1 , x2 , . . . , xm−1 }. Then    xs = (x + cxm )s x∈W  c∈Fp

x∈W

=

s     s j x (cxm )s−j j  j=0

x∈W c∈Fp

166

9 The Transfer

=

=

s    s s−j  s−j  j x c x j m  j=0

j=1

c∈Fp

x∈W

c∈Fp

x∈W

 s s−j  s−j  j c x x j m 

s−1  

   since c∈Fp c0 = 0 and x∈W  x0 = 0. This shows that x∈W xs is nonzero only if there is some value ofj with 1 ≤ j ≤ s − 1 such that   if and s−j j j c∈Fp c x∈W  x = 0. By induction, x∈W  x = 0 unless p − 1 divides  s−j j. By the proof of the m = 1 case, c∈Fp c = 0 unless p − 1 divides s − j. Thus we may concentrate on the values of s and j which are both divisible by p − 1.    s−j j t Suppose sj c∈Fp c x∈W  x = 0 and let s = a0 + a1 p + · · · + at p t and  j = b0 + b1 p + · · · + bt p be the p-adic expansions of s and j. Since s j = 0, by Lucas’ Lemma, we must have br ≤ ar for all r = 0, 1, . . . , t. Since  j < s, there is some value of r with br < ar . Therefore, αp (j) < αp (s). Since x∈W  xj = 0, by induction, we must have αp (j) ≥ (m − 1)(p − 1) and therefore, αp (s) > (m − 1)(p − 1). Since pr ≡ 1 (mod p − 1) for all r, we see that αp (i) ≡ i (mod p − 1) for all non-negative integers i. Therefore, αp (s) ≡ 0 (mod p − 1) and αp (s) > (m − 1)(p − 1) which implies that αp (s) ≥ m(p − 1). Now we prove 2. Take W0 ⊂ W \ {0} such that W0 contains exactly one element of every one dimensional Fp subspace of W (thus |W0 | = (pm −1)/(p− 1)). Thus every non-zero element x ∈ W can be written uniquely as x = cx0

where c ∈ Fp \ {0} and x0 ∈ W0 and d(W ) = x0 ∈W0 ( c∈Fp \{0} cx0 ) =

(−1)|W0 | x0 ∈W0 x0 = (−1)m x0 ∈W0 x0 .   s s The above discussion shows that xp−1 divides  m x∈W x . Since x∈W x p−1 s is a GL(W )-invariant, this shows that y divides x∈W x for all y ∈ W . F[V ] is a Since the elements of W0 are pairwise relatively

prime and since  s unique factorization domain, this shows that x divides 0 x0 ∈W0 x∈W x .  s Therefore, d(W ) divides x∈W x .  m To complete the proof, we consider x∈W xp −1 and d(W ) as polynomials order on F[W ] with in xm with coefficients in F[W  ]. We use  the lexicographic m x1 < x2 < · · · < xm and compare LT( x∈W xp −1 ) and LT(d(W )). The least value of j such that p − 1 divides αp (j) ≥ (m − 1)(p − 1) is j = pm−1 − 1.  j and pm −1 is Therefore, the lead term of x∈W x  m−1  p − 1  pm −pm−1  pm−1 −1 pm −pm−1 c ( x )xm . pm−1 − 1  x∈W

c∈Fp

By induction,



m−1

x∈W 

xp

−1

= d(W  ) and thus the lead term of  m xp −1 x∈W

9.1 The Transfer for Nakajima Groups

is

−d(W  )xpm

m

−pm−1

167

.

On the other hand, LT(d(W )) = LT(



x) = LT((



= d(W  )(



x))

x∈W  \{0}



x)d(W  ) = d(W  )

x∈W \W 



x)(

x∈W \W 

x∈W \{0}

= LT(



LT(x)

x∈W \W  m−1

cxm )p

p−1 p = d(W  )(−xm )

m−1

c∈Fp 

= −d(W )xpm −p .  m Therefore, d(W ) and x∈W xp −1 have the same degree and the same lead term. Since the former divides the latter, they must be equal as claimed.   m

m−1

Proposition 9.1.6. We take P to be a Nakajima group with Nakajima basis {x1 , x2 , . . . , xn } for V ∗ . Then 1. The set Wi := {(σ − 1)xi | σ ∈ P } is an Fp vector space.

n 2. Im TrP is the principal ideal of F[V ]P generated by i=1 d(Wi ). Proof. Let Pi := {σ ∈ P | σ(xj ) = xj for all j = i}. Since P is a Nakajima group , P = Pn Pn−1 . . . P1 and the P -orbit of xi is the same as the Pi -orbit of xi . Since each non-identity element of Pi is a transvection, every such element has order p by Lemma 8.0.3 and thus Pi is an elementary Abelian p-group. Let σ, τ ∈ Pi and write σ(xi ) = xi + yσ and τ (xi ) = xi + yτ where yσ , yτ ∈ span{x1 , x2 , . . . , xi−1 }. Then στ (xi ) = σ(τ (xi )) = σ(xi + yτ ) = σ(xi ) + σ(yτ ) = xi + yσ + yτ and thus (στ − 1)xi = (σ − 1)xi + (τ − 1)xi . This shows that Wi is closed under addition and is therefore an Fp -vector space. For the second statement, let di denote the Fp dimension of Wi . Then F[V ]P = F[N(x1 ), N(x2 ), . . . , N(xn )] where deg(Ni ) = pdi . By Lemma 6.2.1, di

the monomial factors of m := ni=1 xip −1 form a block  basis for F[V ] over F[V ]P . Thus if f ∈ F[V ], we may write (uniquely) f = α divides m fα α where each fα ∈ F[V ]P . Hence   TrP (fα α) = fα TrP (α). TrP (f ) = α divides m

α divides m

Thus to find Im TrP , it suffices to compute TrP (α) for all monomials α which di

divide m = ni=1 xpi −1 . Suppose xA = xa1 1 xa2 2 · · · xann divides m, i.e., suppose ai ≤ pdi − 1 for all i = 1, 2, . . . , n. Since σi (xi ) ∈ span{x1 , x2 , . . . , xi } and (σi − 1)xj = 0 for σi ∈ Pi and i = j, we see that

168

9 The Transfer

    (σn σn−1 · · · σ1 ) xA = (σn σn−1 · · · σ2 ) σ1 (xa1 1 )xa2 2 · · · xann   = (σn σn−1 · · · σ3 ) σ1 (xa1 1 )σ2 (xa2 2 )xa3 3 · · · xann   = (σn σn−1 · · · σ4 ) σ1 (xa1 1 )σ2 (xa2 2 )σ3 (xa3 3 )xa4 4 · · · xann .. . = σ1 (xa1 1 )σ2 (xa2 2 ) · · · σn (xann ) . Therefore, TrP (xa1 1 xa2 2 · · · xann ) =



σ(xA )

σ∈P





=

···

σn ∈Pn σn−1 ∈Pn−1



=



=

=

(σn σn−1 · · · σ1 )(xA )

σ1 ∈P1

···

σn ∈Pn σn−1 ∈Pn−1 n  

 

σ1 (xa1 1 )σ2 (xa2 2 ) · · · σn (xann )

σ1 ∈P1

σi (xi )ai

i=1 σi ∈Pi n  

(xi + y)ai .

i=1 y∈Wi

Now,  y∈Wi

(xi + y)ai =

ai   ai     ai ai −j j  ai ai −j  j xi xi y = y . j j j=0 j=0

y∈Wi

y∈Wi

 By Lemma 9.1.5, y∈Wi y j equals 0 unless p−1 divides j and αp (j) ≥ di (p−1). Since j ≤ ai < di for all i = 1, 2, . . . , n we have, TrP (xa1 1 xa2 2 · · · xann ) = 0 unless ai = di − 1 for all i = 1, 2, . . . , n. By the second statement in Lemma 9.1.5, we have TrP (x1d1 −1 xd22 −1 · · · xndn −1 ) =

n  

(xi + y)di −1

i=1 y∈Wi

=

n  

y di −1

i=1 y∈Wi

=

n  i=1

d(Wi ).  

We remind the reader that for V = Fn , we showed in §3.4 that the ring of invariants F[V ]Un is the polynomial ring on the invariants h1 , h2 , . . . , hn .

9.1 The Transfer for Nakajima Groups

169

Lemma 9.1.7. We have that Im TrUn (Fq ) is the principal ideal generated by hn−2 · · · hn−1 )q−1 . (hn−1 1 2

n Proof. By Proposition 9.1.6, Im TrUn (Fq ) is generated by i=1 d(Wi ) where Wi is the Fq vector space spanned by {x1 , x2 , . . . , xi−1 }. We will show that d(Wt ) = −hq−1 t−1 d(Wt−1 ) for all t = 1, 2, . . . , n. By convention, we take d(W0 ) = −1 and h0 = 1. We proceed by induction on t. d(W0 ). For t = 1, we have W1 = {0} and thus d(W1 ) = 1 = hq−1 0 Assume then that d(Wt ) = hq−1 d(W ) and consider d(Wt+1 ). We have t−1 t−1 d(Wt+1 ) =

 x∈Wt+1 \{0}

=



x=





x

x∈Wt+1 \Wt

y

y∈Wt \{0}

xd(Wt )

x∈Wt+1 \Wt





=d(Wt )

cxt+1 + z

c∈Fq \{0} z∈Wt



=d(Wt )



c(xt+1 + z/c)

c∈Fq \{0} z∈Wt





=d(Wt )

c

c∈Fq \{0}

⎛

=d(Wt ) ⎝



c∈Fq \{0}

=d(Wt )(−1)





 xt+1 + z



z  ∈Wt

⎞ c⎠



q−1 xt+1 + z



z  ∈Wt

(xt+1 + z  )q−1 = −d(Wt )ht .

z  ∈Wt

From this get d(Wt ) = (−1)t (h1 h2 · · · ht−1 )q−1 . Thus Im TrUn (Fq ) is gener we n ated by t=1 d(Wt ) = ± nt=1 (h1 h2 · · · ht−1 )q−1 = ±(hn−1 hn−2 · · · hn−1 )q−1 . 1 2   Corollary 9.1.8. Let V be an n dimensional vector space over Fq . Then Im TrGL(V ) is the principal ideal of F[V ]GL(V ) generated by d(V )n−1 . Proof. Suppose that f ∈ Im TrGL(V ) . By Proposition 9.0.18, Im TrGL(V ) = Im TrUn (Fq ) ∩Fq [V ]GL(V ) . By the preceding lemma, Im TrUn is the principle ideal generated by (hn−1 hn−2 · · · hn−1 )q−1 . 1 2 Therefore, hn−2 · · · hn−1 )q−1 h f = (hn−1 1 2

170

9 The Transfer

q−1 for some h ∈ Fq [V ]Un (Fq ) . Since and d(V ) = c∈Fq \{0} cx1 = −x1

q−1 q y, we see that x divides d(V ) but x does not divide d(V ). 1 1 y∈V ∗ \{0} (n−1)(q−1)

Since h1 = x1 , we see that x1 divides f . Take any non-zero element y ∈ V ∗ . Since x1 and y lie in the same GL(V )-orbit, and since f is GL(V )invariant, we see that y (n−1)(q−1) also divides f . Putting these facts together we get that d(V )n−1 divides f . For the opposite inclusion, we must show that d(V )n−1 lies in Im TrGL(V ) . But d(V ) = (−1)n (h1 h2 . . . hn )q−1 and thus using the preceding lemma, d(V )n−1 lies in Im TrUn (Fq ) . But clearly, from its definition, d(V ) ∈ Fq [V ]GL(V ) . Therefore, d(V )n−1 ∈ Im TrUn (Fq ) ∩Fq [V ]GL(V ) = Im TrGL(V ) .  

9.2 Cohen-Macaulay Invariant Rings of p-Groups Hochster and Eagon [53] have shown that non-modular invariant rings are always Cohen-Macaulay. A very important question in modular invariant theory is whether or not the invariant ring is Cohen-Macaulay. Campbell, Hughes and Pollack [18] show that a sufficient condition is that the p-Sylow subgroup have a Cohen-Macaulay invariant ring. In this section, we give a necessary and restrictive condition for the invariant ring of a p-group to be Cohen-Macaulay, Theorem 9.2.2. This result is the culmination of a sequence of papers beginning with Ellingsrud and Skjelbred [35] and leading to Campbell, Geramita, Hughes, Shank, Wehlau [17], and Kemper [65]. In particular, this result can be viewed as a generalization of the calculation given in Example 4.0.4. The proof we give here of Theorem 9.2.2 avoids the use of group cohomology. Definition 9.2.1. A element σ ∈ GL(V ) is called a bi-reflection if the endomorphism σ − IV of V has rank less than or equal to 2. Here σ − IV is a linear endomorphism of V and thus its rank is just the rank of the corresponding matrix. Thus σ is a bi-reflection if and only if dimF (V σ ) ≥ dimF (V ) − 2. Kemper’s proof of the following result uses group cohomology. Theorem 9.2.2. Let G ≤ GL(V ) be a p-group and suppose that F[V ]G is a Cohen-Macaulay ring. Then G is generated by bi-reflections. Proof. Suppose that F[V ]G is Cohen-Macaulay. Let H be the subgroup of G generated by all the bi-reflections in G. Assume, by way of contradiction, that

9.2 Cohen-Macaulay Invariant Rings of p-Groups

171

H = G. By Lemma 1.10.3, there exists a maximal proper subgroup N of G such that H ≤ N , N G and [G : N ] = p. Take σ ∈ G\N . By Theorem 9.0.10, σ

N IF[V ]G (V V (TrG N (F[V ] )) = IF[V ]G (∪order(σN )=p V )

and thus

 N TrG N (F[V ] ) =



σ

IF[V ]G (V ) .

order(σN )=p

Here we are writing IF[V ]G (X) for X ⊆ V to denote the ideal of functions in F[V ]G which vanish on πV,G (X) ⊂ V //G, i.e., IF[V ]G (X) denotes IF[V ]G (πV,G (X)) = IF[V ] (X) ∩ F[V ]G . σ σ Now IF[V ]G (V ) = F[V ]G ∩ IF[V ] (V ). Therefore by the going-up and going-down Theorem 2.5.2 (2) and 2.5.2 (3), σ

σ

height(IF[V ]G (V )) = height(IF[V ] (V )). σ

σ

But V is just a subspace of V and clearly height(IF[V ] (V )) = rank(σ − IV ). Since an ideal and its radical have the same height, we see that N height(TrG N (F[V ] )) = min{rank(σ − IV ) | order(σN ) = p} ≥ 3, by definition of H and N .

But if R is any Noetherian ring and I is an ideal of R of height m, then (by [79][Theroem 24 14.F] say) there exist a1 , a2 , . . . , am ∈ I such that the ideal (a1 , a2 , . . . , am ) also has height m. Therefore, there exist a1 , a2 , a3 ∈ N TrG N (F[V ] ) such that a1 , a2 , a3 is a partial homogeneous system of parameters G in F[V ] . Since F[V ]G is Cohen-Macaulay, a1 , a2 , a3 is a regular sequence in F[V ]G . N Write ai = TrG for i = 1, 2, 3. Now the map TrG N (fi ) with fi ∈ F[V ] N = 2 p−1 p−1 = (σ − 1) and thus we have ai = (σ − 1)p−1 (fi ). 1 + σ + σ + ··· + σ Define bi by bi := (σ − 1)p−2 (fi ) (so bi = fi if p = 2) for i = 1, 2, 3. Then (σ − 1)(bi ) = ai . Also, note that for any h ∈ F[V ]N and any n ∈ N , we have n · (σ − 1)(h) = nσ(h) − nh = σn (h) − h = (σ − 1)h where n = σ −1 nσ ∈ N . Thus bi = (σ − 2)p−1 (fi ) ∈ F[V ]N for all i = 1, 2, 3. Expanding ⎛⎛ ⎞⎞ a1 a2 a3 det ⎝⎝a1 a2 a3 ⎠⎠ = 0 b1 b2 b3 along the first row, we have a1 (a2 b3 −a3 b2 )+a2 (a3 b1 −a1 b3 ) = a3 (a2 b1 −a1 b2 ). It is easily verified that (σ − 1)(ai bj − aj bi ) = 0 and thus ai bj − aj bi ∈ F[V ]G for all 1 ≤ i = j ≤ 3. Therefore, by the definition of a regular sequence, there must exist h1 , h2 ∈ F[V ]G such that a2 b1 − a1 b2 = h1 a1 + h2 a2 . Thus a2 (b1 − h2 ) = a1 (h1 + b2 ). Again, by the definition of a regular sequence, this implies that there exist t ∈ F[V ]G such that b1 −h2 = ta1 and thus b1 ∈ F[V ]G . But then a1 = (σ − 1)b1 = 0. This contradiction shows that H must be equal to G, i.e., that G is generated by its bi-reflections.  

172

9 The Transfer

More general versions of this result and related material may be found in Kemper [65, §3], and Lorenz [77, §8]. Corollary 9.2.3. Let V be a faithful modular representation of a p group G defined over a field F. Then F[3 V ]G is not Cohen-Macaulay. Proof. Consider e = σ ∈ G. Then dim V σ ≤ dim V −1. Therefore dim(3 V )σ ≤ 3(dim V − 1) = dim(3 V ) − 3. Thus no non-trivial element of G can act on 3 V as a bi-reflection.   Example 4.0.4 is a simple example of Corollary 9.2.3. Historically it was study of this example which lead to the development of the proof of Theorem 9.2.2. Theorem 3.9.2 provides a partial converse to Theorem 9.2.2, by giving a class of representations for which all elements are bi-reflections and for which the ring of invariants is always Cohen-Macaulay. Note that the converse of Theorem 9.2.2 is not valid. For example, at the end of §1.1.1, we exhibited an example of a group generated by reflections whose ring of invariants is not Cohen-Macaulay. Let H be a subgroup of G such that H contains a p-Sylow subgroup of G. We have already seen in Proposition 9.0.18 that this condition means there is closer relation between F[V ]H and F[V ]G than in general. The next result shows there is another similarity between these two rings. Proposition 9.2.4. Suppose that H is a subgroup of G such that [G : H] is invertible in F. If F[V ]H is Cohen-Macaulay, then F[V ]G is also CohenMacaulay. Proof. Choose a homogeneous system of parameters f1 , f2 , . . . , fn for F[V ]G . Then by Corollary 3.0.6, this is also a homogeneous system of parameters for F[V ]H and F[V ]. Let r denote the index of H in G. Recall that 1/r TrG H : F[V ]H → F[V ]G is the Reynolds operator and accordingly we have the F[V ]G module decomposition F[V ]H = F[V ]G ⊕ U where U := ker TrG H. Let B := F[f1 , f2 , . . . , fn ] and let I denote the ideal of F[V ]H generated by (f1 , f2 , . . . , fn ) and let J := I ∩ F[V ]G . The above decomposition implies that F[V ]H /I ∼ = F[V ]G /J ⊕ U/K where K = f1 U + f2 U + · · · + fn U . Choose a basis {h1 , h2 , . . . , hr } for F[V ]G /J and another basis {hr+1 , hr+2 , . . . , hs } for U/K. Then {h1 , h2 , . . . , hs } is a basis for F[V ]H /I. Lift each hi to an element hi ∈ F[V ]H . Then

9.3 Differents in Modular Invariant Theory

F[V ]H =

s 

173

Bhi

i=1

r and F[V ]G = i=1 Bhi . But since F[V ]H is Cohen-Macaulay, we have in fact that F[V ]H = ⊕si=1 Bhi and F[V ]G = ⊕ri=1 Bhi . Thus, since it is a free Bmodule, F[V ]G is Cohen-Macaulay.   Remark 9.2.5. Let V be a fixed non-trivial modular representation of the group G. Then Gregor Kemper has proved that for sufficiently large m, the ring of vector invariants F[mV ]G is not Cohen-Macaulay. See [65]. It remains an open question whether m = 3 is sufficiently large. Suppose G is a p-group generated by reflections and F[V ]G is Cohen-Macaulay. There are examples that show that sometimes, but not always, F[2V ]G is Cohen-Macaulay. It is an open problem to characterize those V for which F[2V ]G is Cohen-Macaulay. See the work of Chuai [23] and Shank and Wehlau [99].

9.3 Differents in Modular Invariant Theory Here we recall work of A. Broer [13], closely following his exposition. His work is best understood in the context of §2.5. That is, suppose that R ⊂ S are connected graded commutative rings with unit, that S is integral over R, both of which are domains and integrally closed in their quotient fields Quot(R) and Quot(S), and that the extension Quot(R) ⊂ Quot(R) is separable. Then the trace map Tr : Quot(S) → Quot(R) is surjective, respects the grading, and Tr(S) ⊆ R. We suppose that, as a R-module, S has homogeneous generators s1 , . . . , sm , and we define an epimorphism of algebras ρ

R[a1 , a2 , . . . , am ] → → S by the rule ρ(a ) = s where deg(a ) = deg(s ) with kernel I. We refer to elements of I as relations in S and to I as the ideal of relations. Now S integral over R implies that there is a monic polynomial  f (t) = rj, tj , where rj, ∈ R and f (s ) = 0 . As before, we write R+ for the ideal of positive degree elements of R and we consider the Hilbert ideal R+ S of S. The quotient algebra S/(R+ S) is a finite dimensional algebra. In the event that R = S G , we call S/(R+ )G the ring of coinvariants of G.

174

9 The Transfer

9.3.1 Construction of the Various Different Ideals The Noether Different Ideal The enveloping algebra of the extension S over R is the algebra S e = S ⊗R S. We define J to be the kernel of the multiplication map μ

S ⊗R S → S s ⊗ s → ss . Then J is generated by s ⊗ 1 − 1 ⊗ s , 1 ≤  ≤ m. We define K to be the annihilator of J in S e and the Noether different ideal as the image μ(K) = N N if and only if there is a t ∈ S e such DN S/R = D ⊂ S. In other words, s ∈ D  that μ(t) = s and for all s ∈ S we have (1 ⊗ s )t = (s ⊗ 1)t. The Galois Different Ideal Consider the natural map associated to an element ω : S ⊗ S → EndR (S) [ω(t ⊗ t )](s) =





t ⊗ t ∈ S e

t Tr(t s) .

Then ω is a S e -module map under the natural S e -module structure on EndR (S), and the image ω(S e ) ⊂ EndR (S) is the cyclic S e -module generated by Tr. Let ε

EndR (S) → S ε(η) = η(1S ) . We define the Galois different ideal to be Gal = {ε(η) | η ∈ ω(S e ) ∩ EndR R} = ε(ω(S e ) ∩ EndR R . DGal S/R = D

We note that EndR R is the collection of maps R → R given by multiplication by  the elements of R, the elements of the Galois different are the maps s → t Tr(t s). There is an alternate description of DGal in the case that R = S G . We consider the twisted group ring (SG, ·) which is additively the free S-module on generators {eσ | σ ∈ G} with multiplication determined by the formula s1 eσ · s2 eτ = s1 σ(s2 )eστ . The unit of the group is the unit of the twisted group ring, and (SG, ·) is naturally a graded module over S e . There is also  a natural inclusion of asα= s eσ to the R-linear map sociative rings (SG, ·) → EndR (S) taking  ρα (s) = s σ (s). Of course, the element σ∈G eσ determines the trace over G. The Galois  different is equal to the intersection of the two-sided ideal generated by σ∈G eσ with the subring S (that is S[IdG ]). Broer notes that an extension S G ⊂ S is said to be Galois if the Galois different ideal is the trivial ideal (1S ), providing a rationale for the language.

9.3 Differents in Modular Invariant Theory

175

The Inverse Dedekind Different Ideal The inverse Dedekind different ideal is defined as −1 = {r ∈ Quot(R) | for all s ∈ S, Tr(rs) ∈ R} (DD )−1 = (DD S/R )

We note that, for every R-linear map η : S → R, there is a unique r ∈ (DD )−1 such that η(s) = Tr(rs) for all s ∈ S. Therefore, we define the Dedekind different ideal to be   D −1 DD = DD ⊂R . S/R = r ∈ Quot(R) | r(D ) Under our hypotheses, the Dedekind different is a divisorial graded ideal in S, that is, the height of any one of its primary components is one. If, in addition, S is a unique factorization domain, then the Dedekind different ideal is a homogeneous principal ideal. In this case, we denote a generator by θ. The Twisted Trace Different Ideal Here we define the twisted trace different ideal or T -different ideal to be DT = DTS/R = DD (Tr((DD )−1 )). When DD = (θ), we define the twisted trace to be the R-module homomorphism given by Tr

S →θ S Trθ (s) = θ Tr(sθ) . In this case, since have that

S θ

−1 = (DD , we do have Trθ (sθ) ∈ θR ⊂ S. Then we S/R )

DD = S Trθ (S). The K¨ ahler Different Ideal Recall that J is defined to be the kernel of the multiplication map S e = ahler differentials is defined S ⊗ S → S. The graded S = S e /J-module of K¨ to be Ω = ΩS/R = J/J 2 . Recall the ideal of relations I determined by ρ

I ⊂ R[a1 , a2 , . . . , am ] → S . We suppose that I is generated by {f1 , f2 , . . . , fk }. For each such f , we define df =

∂f ∂f ∂f dt1 + dt2 + · · · + dtm ∂t1 ∂t2 ∂tm

176

9 The Transfer

where

∂f ∂f = μ( ). ∂tj ∂aj

Then Ω is generated by the dt = t ⊗ 1 − 1 ⊗ t + J 2 , 1 ≤  ≤ m subject to the relations just given and we obtain a presentation of Ω by free graded S-modules F1 → F0 → ΩS/R where F1 is of rank k and F0 is of rank m. The S-linear map between F1 and F0 is associated to the m × k matrix A with entries Ai,j =

∂fj . ∂ti

The K¨ ahler different ideal or Jacobian ideal, denoted DK = DK S/R , is the annihilator AnnS (Ω), in other words, the zero-th Fitting ideal of Ω. We note that DK is generated by the determinants of all m × m-minors of A. The elements of this theory show that the K¨ ahler different ideal does not depend upon the choices of presentation or homogeneous bases. For any m relations {f1 , f2 , . . . , fm } ⊂ R[a1 , a2 , . . . , am ], the Jacobian determinant is defined to be Jac(f1 , f2 , . . . , fm ) = det(

∂fj )∈S ∂ti

and the K¨ ahler  k  different ideal is the ideal generated by the Jacobians associm-tuples of generators of I. ated to the m Now suppose that S = F[x1 , x2 , . . . , xn ] is a graded polynomial algebra on algebraically independent generators x . We consider the presentation of R-algebras given by ρ I → R[a1 , a2 , . . . , an ] → S given by ρ(a ) = x . Each element r ∈ R can be written as a polynomial in the x ’s, that is, r = r(x1 , x2 , . . . , xn ), and we can construct an element r(a) = r(a1 , a2 , . . . , an ) ∈ R[a1 , a2 , . . . , an ]. On the other hand, we can simply consider r as a constant in R[a1 , a2 , . . . , an ]. Then the difference r(a) − r ∈ I is a relation. Therefore, for any 1 ≤  ≤ n we have ∂(r(a) − r) ∂(r(a)) ∂(r(x)) ∂(r(a) − r) = μ( ) = μ( )= . ∂x ∂a ∂a ∂x This latter expression is the usual partial derivative of R as an element of F[x1 , x2 , . . . , xn ]. It follows in this case that Jac(

∂(r(a) − r) ∂(r(a)) ) = Jac( ), ∂x ∂x

the classical Jacobian. Further, since the relations r(a) − r generate I, we see that DK is generated by the Jacobians of the m-tuples of elements from R.

9.3 Differents in Modular Invariant Theory

177

Differents and Properties of Modular Invariant Rings Broer has determined the inclusions of the various differents as follows. Theorem 9.3.1. 1. We have DK ⊆ DN ⊆ DGal ⊆ DD and DN ⊆ Ann Ω; 2. We have DT ⊂ DD and if DT is a principal ideal, then DGal ⊆ DT ; 3. If S is free as a graded R-module, then DN = DGal = DT = DD .   Broer’s techniques allow him to analyze how various properties of modular invariant rings are related to properties of the differents. We recall here two of his theorems. First, recall that S/(R+ ) is said to be a Poincar´e duality algebra if there exists a non-degenerate symmetric bilinear mapping S/(R+ ) ⊗ S/(R+ ) → S/(R+ )

Theorem 9.3.2. The following statements for R ⊂ S are equivalent: 1. S/(R+ ) is a Poincar´e duality algebra; 2. DN is a principal ideal; 3. DGal is a principal ideal; 4. DD is a principal ideal and DN = DGal = DD .   In the case of most interest for this chapter, Broer proves Theorem 9.3.3. We assume that S = F[x1 , . . . , xn ] is a graded polynomial algebra with deg(x ) > 0 for all  = 1, 2, . . . , n. We consider the extension R ⊂ S as in this section. Then the Dedekind different DD = (θ) is principal, and the following statements are equivalent. 1. The ring R is also a polynomial algebra; 2. S is free as graded R-module; ´ algebra; 3. S/(R+ ) is a Poincarduality K N 4. Either D or D or DGal equals DD ; D 5. We have DK = DN = DGal = DT = D ; deg(f ) − deg(x ) = deg(θ) and 6. There exists {f1 , f2 , . . . , fn } such that 0 = Jac(f1 , f2 , . . . , fn ); 7. There exists {f1 , f2 , . . . , fn } such that R = F[f1 , f2 , . . . , fn ].  

10 Invariant Rings via Localization

In some circumstances, it is possible to study a ring of invariants through a well-chosen localization. In this chapter, we give some instances of this technique. Let R be a finitely generated algebra which is a domain. Recall that Rf denotes the localization of R with respect to the multiplicative set generated by f . Proposition 10.0.1. Suppose that B is a subalgebra of R and that f1 , f2 is a regular sequence in B such that Bf1 = Rf1 and Bf2 = Rf2 . Then B = R. Proof. Take h ∈ R. Since R ⊆ Rfi = Bfi we may write h = b1 /f1n and h = b2 /f2m for some b1 , b2 ∈ B and n, m ∈ N. Therefore, b1 f2m = b2 f1n . Since f1 , f2 is a regular sequence in B, so also is f1n , f2m . This implies that b2 = bf2m for some b ∈ B. Thus h = bf2m /f2m = b lies in B.   Lemma 10.0.2. Suppose that B is a subalgebra of R and I is an ideal of R. √ If I ⊆ B and f ∈ I ∩ B, then Bf = Rf . Proof. There exists m ∈ N such that f m ∈ I. Take h ∈ Rf and write h = r/f k with r ∈ R. Then rf m = hf k+m ∈ I ⊆ B and h = rf m /f k+m ∈ Bf .   G Theorem 10.0.3. Suppose B is a subalgebra  of F[V ] containing the image of the transfer, Im TrG ⊂ B. Let f1 , f2 ∈ Im TrG ∩ B be two elements such that f1 , f2 is a regular sequence in B. Then B = F[V ]G .

Remark 10.0.4. Suppose that B is a subalgebra of F[V ]G containing the image of the transfer.  Further, suppose that f1 and f2 are non-associate primes of B lying in Im TrG . A relatively routine calculation shows that f1 , f2 is a regular sequence in B and so B = F[V ]G . Remark 10.0.5. Let F denote the algebraic  closure of F and consider V := F ⊗ V . By Theorem 9.0.10, we have that Im TrG consists of those invariant H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, 179 Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 10, © Springer-Verlag Berlin Heidelberg 2011

180

10 Invariant Rings via Localization

polynomials in F[V ] which vanish on the subvariety V of V defined by V = σ ∪σ∈Σ V where Σ consists of all the elements σof G of order p. Thus given an element f ∈ F[V ]G , we may check that f ∈ Im TrG by verifying that f σ vanishes on V for every element σ ∈ G of order p. We illustrate Theorem 10.0.3 by using it to compute (again) the ring of Cp invariants, F[2 V2 ]Cp . Let σ be a generator for Cp . Choose a triangular basis y2 , x2 , y1 , x1 for (2 V2 )∗ with σ(yi ) = yi + xi and σ(xi ) = xi for i = 1, 2. We use the graded reverse lexicographic order on F[2 V2 ] with x1 < y1 < x2 < y2 . We take Ni = N(yi ) = yip − xip−1 yi for i = 1, 2 and u = x2 y1 − x1 y2 . Let B denote the ring B = F[x1 , x2 , u, N1 , N2 ]. We will use Theorem 10.0.3 to show that F[2 V2 ]Cp = B. We begin by verifying that Im TrCp ⊂ B. Lemma 10.0.6. Let 0 ≤ a, b ≤ p − 1 and put e := a + b − (p − 1). ⎧ ⎪ if e < 0; ⎨0, TrCp (y1a y2b ) = −up−1 − (x1 x2 )p−1 , if a = b = p − 1; ⎪ ⎩ a p−1−b p−1−a e x2 u , otherwise. − e x1 Proof. TrCp (y1a y2b ) =

p−1

(y1 + kx1 )a (y2 + kx2 )b

k=0

=

p−1 b  a  a i i a−i b j j b−j k x1 y1 k x2 y 2 i j i=0 j=0 k=0

=

b   a a b i=0 j=0

i

j

xi1 y1a−i xj2 y2b−j

p−1

k i+j .

k=0

p−1 By Lemma 9.0.2, k=0 k i+j is zero unless i + j = p − 1 or i + j = 2(p − 1) in which case this sum is -1. Therefore, TrCp (y1a y2b ) = 0 if a+b < p−1. Moreover, if p − 1 ≤ a + b < 2(p − 1), then Cp

Tr

(y1a y2b )

 b =− xi1 y1a−i x2p−1−i y2b−p+1+i p − 1 − i i i=0 



a b a =− xi y a−i x2p−1−i y2b−p+1+i p−1−i 1 1 i a 

a

i=p−1−b

=−

a+b−p+1

t=0



a p−1−b+t

 b t xp−1−b+t y1a+b−p+1−t xb−t 2 y2 b−t 1

10 Invariant Rings via Localization

=

−x1p−1−b xp−a−1 2

a+b−p+1

t=0



a p−1−b+t

 b b−t

181



xt1 y1a+b−p+1−t xb−t−p+1+a y2t 2 = −x1p−1−b xp−a−1 2

e

t=0



a p−1−b+t

 b t xt y e−t xe−t 2 y2 . b−t 1 1

Wilson’s asserts that (p − 1)! = −1 (mod p) and Lucas’ Lemma   Theorem s = (−1) (mod p). Using these two facts we get gives p−1 s  

a! b! b a = (p − 1 − b + t)! (a + b − p + 1 − t)! t! (b − t)! t p−1−b+t a! b! = (e − t)! t! (p − 1 − b + t)! (b − t)! (p − 1)! a! b! =− (e − t)! t! (p − 1 − b + t)! (b − t)! a! b! =− (−1)b−t (e − t)! t! a! b! =− (−1)b (−1)t (e − t)! t! a! b! (p − 1)! = −(−1)t (e − t) !t! (p − 1 − b)! b! a! b! 1 = (−1)t (e − t)! t! (p − 1 − b)! b! b! a! = (−1)t (p − 1 − b)! b!(e − t)! t! a! e! = (−1)t (p − 1 − b)! e! (e − t)! t!

  a e = (−1)t . t e Therefore, for p − 1 ≤ a + b < 2(p − 1) we have Tr

Cp

(y1a y2b )

=

−x1p−1−b xp−a−1 2

 e t e−t e−t t (−1) x y x2 y 2 t 1 1 t=0

e

t

 a p−1−b p−a−1 x =− x2 (x2 y1 − x1 y2 )e e 1

 a p−1−b p−a−1 e x =− x2 u . e 1

For a = b = p − 1, we get one extra term corresponding to i + j = 2(p − 1)   which does not vanish. Thus TrCp (y1p−1 y2p−1 ) = −up−1 − (x1 x2 )p−1 .

182

10 Invariant Rings via Localization

Using Lemma 10.0.6, we can see that TrCp (F[2 V2 ]) ⊂ F[x1 , x2 , u, N1 , N2 ] as follows. The four invariants x1 , x2 , N1 , N2 form a homogeneous system of parameters for F[2 V2 ]. Taking H = F[x1 , x2 , N1 , N2 ] we have the Hironaka p−1 p−1 decomposition F[2 V2 ] = ⊕a=0 ⊕b=0 Hy1a y2b . Since TrCp is H-linear, we see p−1 p−1 Cp a b that Tr (F[2 V2 ]) ⊂ a=0 b=0 H · TrCp (y 1 y2 ) ⊂ F[x1 , x2 , u, N1 , N2 ].

Lemma 10.0.6 also shows that x1 , x2 ∈ Im TrCp since Tr(y1p−1 ) = x1p−1 and Tr(y2p−1 ) = x2p−1 . To use Theorem 10.0.3 to show that F[2 V2 ]Cp = F[x1 , x2 , u, N1 , N2 ], it only remains to show that x2 , x1 is a regular sequence in B = F[x1 , x2 , u, N1 , N2 ]. We begin by noting that x1 , x2 , u, N1 , N2 is a SAGBI basis for B. This is clear since LT(x1 ) = x1 , LT(x2 ) = x2 , LT(u) = x2 y1 , LT(N1 ) = y1p and LT(N2 ) = y2p . Thus the only non-trivial tˆete-a-tˆete difference is up − xp2 N1 which subducts to zero via up − xp2 N1 + xp1 N2 − (x1 x2 )p−1 u = 0. In fact, we already found this SAGBI basis in Example 5.1.9. To see that x2 , x1 is a regular sequence in B, we suppose that x1 g = x2 h for some g, h ∈ B. Then x1 divides LT(h). We need to show that x1 divides h in B. This follows from the following lemma. Lemma 10.0.7. Suppose B is a SAGBI basis for a ring B ⊂ F[x1 , x2 , . . . , xn ] with respect to a graded reverse lexicographic term order with x1 < xi for all i = 2, 3, . . . , n. Further, suppose that x1 ∈ B and that x1 is the unique element of B whose lead term is divisible by x1 . Let f ∈ B. If x1 divides LT(f ), then x1 divides f in B. Proof. Assume that the result is false. Then there exist elements f ∈ B whose lead term is divisible by x1 but for which f /x1 ∈ / B. Choose such an f with minimal lead term. Since B is a SAGBI basis for B, we may write LT(f ) = LT(x1 )e0 LT(g1 )e1 LT(g2 )e2 · · · LT(gs )es where x1 , g1 , g2 , . . . , gs ∈ B and e1 , e2 , . . . , es are non-negative integers. Since x1 divides LT(f ) and x1 does not divide LT(gi ) for i = 1, 2, . . . , s, we must have e0 ≥ 1. Define f  = f − xe10 g1e1 g2e2 · · · gses . Then LT(f  ) < LT(f ). Since we are using a graded reverse lexicographic ordering with x1 small, this implies that x1 divides LT(f  ). Then by the minimality of f we see that h = f  /x1 ∈ B. But then f = x1 (xe10 −1 g1e1 g2e2 · · · gses + h ) is divisible by x1 in B.   An important application of the above lemma is in combination with the following proposition. Proposition 10.0.8. Let A = F[f1 , f2 , . . . , fr ] ⊆ F[V ]G . Suppose 1. F[V ] is integral over A; 2. Quot(A) = F(V )G ; 3. There exists h ∈ A such that hA is a prime ideal of A and Ah is a unique factorization domain. Then A = F[V ]G .

10 Invariant Rings via Localization

183

Proof. By [6, Lemma 6.3.1, p. 73], hypothesis (3) implies that A is itself a unique factorization domain and is therefore by Proposition 3.0.2 integrally closed. Let f ∈ F[V ]G be arbitrary. By (2), f lies in Quot(A). Then (1) implies that f is integral over A and since A integrally closed, f ∈ A. Therefore   A = F[V ]G . Verifying the first hypotheses of Proposition 10.0.8 is usually easy as all we need to do is ensure that A contains a homogeneous system of parameters. Verifying the second hypothesis is a little harder but the following result due to Kemper, [64, Cor 1.8], is often applicable or by the use of Theorem 4.3.4. Proposition 10.0.9. Let f1 , f2 , . . . , fn be n algebraically independent homogeneous polynomials over a field K. Then the  degree of the field extension [K(x1 , x2 , . . . , xn ) : K(f1 , f2 , . . . , fn )] is at most i=1 deg(fi ). G Corollary 10.0.10. Let f1 , f be algebraically independent 2 , . . . , fn ∈ K[V ] homogeneous invariants. If i=1 deg(fi ) < 2|G|, then K(f1 , f2 , . . . , fn ) = K(V )G .

Proof. Since K(f1 , f2 , . . . , fn ) ⊆ K(V )G ⊆ K(x1 , x2 , . . . , xn ), we have  2|G| > deg(fi ) i=1

≥ [K(x1 , x2 , . . . , xn ) : K(f1 , f2 , . . . , fn )] = [K(x1 , x2 , . . . , xn ) : K(V )G ] · [K(V )G : K(f1 , f2 , . . . , fn )] = |G| · [K(V )G : K(f1 , f2 , . . . , fn )]. Thus [K(V )G : K(f1 , f2 , . . . , fn )] < 2.

 

Verifying the third hypothesis of Proposition 10.0.8 is generally much harder than verifying the first two hypotheses. In particular, it is hard to show that the element h is prime in the ring A which is usually given by specifying a set of generators for A. One way to do this however is when the invariant h is x1 . If a SAGBI basis for A with respect to grevlex, with x1 small, is known, then Lemma 10.0.7 can be used to show that the principal ideal x1 A is a prime ideal of A. We illustrate this with a simple example. Example 10.0.11. We return to the 4 dimensional representation of Cp 3 considered in Section 8.2. This representation is given by ⎧⎛ ⎫ ⎞ 1000 ⎪ ⎪ ⎪ ⎪ ⎨⎜ ⎬ 0 1 0 0⎟ ⎜ ⎟ G= ⎝ . | a, b, c ∈ F p a c 1 0⎠ ⎪ ⎪ ⎪ ⎪ ⎭ ⎩ cb01 Let {x1 , x2 , y1 , y2 } be the basis of V ∗ dual to the standard basis of V . We will use the graded reverse lexicographic ordering with x1 < x2 < y1 < y2 .

184

10 Invariant Rings via Localization

One may verify that the following functions are G-invariants: x1 x2 h = x2 (y2p − x2p−1 y2 ) + x1 (y1p − x1p−1 y1 ) = x2 y2p + . . . 2

p f 1 = NG Gy1 (y1 ) = y1 + . . . 2

p f2 = NG Gy2 (y2 ) = y2 + . . .

We will show that F[V ]G = F[x1 , x2 , h, f1 , f2 ] by applying Lemma 10.0.7 and Proposition 10.0.8. We seek a SAGBI basis for F[V ]G . Accordingly, we consider the lone tˆete-a-tˆete among x1 , x2 , h, f1 , f2 which is hp − xp2 f2 . We perform the subduction of this tˆete-a-tˆete and obtain hp − xp2 f2 − xp1 f1 − h

p

(x1p−j+1 xj2 )p−1 = 0 .

j=1

Thus {x1 , x2 , h, f1 , f2 } is a SAGBI basis for the algebra A := F[x1 , x2 , h, f1 , f2 ] 2 which they generate. Since LT(x1 ) = x1 , LT(x2 ) = x2 , LT(f1 ) = y1p and 2 LT(f2 ) = y2p , we see by Lemma 6.2.1 that x1 , x2 , f1 , f2 forms a homogeneous system of parameters and so F[V ] is integral over A. Since the lead terms of x1 , x2 , h, f1 are algebraically independent, so are these 4 polynomials themselves. Applying Corollary 10.0.10 we see that F(x1 , x2 , h, f1 ) = F(V )G . By Lemma 10.0.7, we know that Ax is a prime ideal of A. Furthermore, from the subduction of the tˆete-a-tˆete above, we see that f1 ∈ F[x1 , x2 , h, f2 ]x1 and therefore, Ax1 = F[x1 , x2 , h, f1 , f2 ]x1 = F[x1 , x2 , h, f2 ]x1. Since F[x1 , x2 , h, f2 ] is a polynomial ring, this implies that Ax1 is a unique factorization domain. Applying Proposition 10.0.8 we see that F[V ]G = A = F[x1 , x2 , h, f1 , f2 ].

11 Rings of Invariants which are Hypersurfaces

As we have seen, we seek to characterize those representations V of groups G whose rings of invariants are well-behaved. The best behaved rings of invariants, K[V ]G , are those which are polynomial rings, that is, K[V ]G is generated by dim(V ) many invariants. A slightly less well behaved class of examples is provided by those rings of invariants which are hypersurfaces, that is, K[V ]G is generated by dim(V ) + 1 many invariants. Those representations with this property have been extensively studied in characteristic 0 by Nakajima [84]. Less is known for modular groups. When G is a Nakajima group with maximal proper subgroup H, the following proposition shows that the ring of H-invariants is a hypersurface (or a polynomial) ring. Proposition 11.0.1. Let R be an integral domain of characteristic p and suppose the finite group G acts faithfully on R. Suppose H ≤ G is a maximal proper subgroup of index k ≤ p. Let σ ∈ G \ H. If there exists y ∈ RH such that x := (σ − 1)y lies in RG and such that the set (σ − 1)RH lies in the principal ideal of R generated by x, then RH = RG [y]. Proof. Since the group H is a proper subgroup of G, the field Quot(R)G is a proper subfield of Quot(R)H (in fact, the degree of the extension Quot(R)G ⊂ Quot(R)H is |G|/|H|) and thus the ring RG is a proper subring of RH . Thus there exists f ∈ RH \ RG . Since G is generated by H together with σ, we see that (σ − 1)f is a non-zero element of R which is divisible by x. This shows that x = 0 and hence that y is not G-invariant. Therefore, y ∈ / Quot(R)G . By the maximality of H in G, this implies, using Galois Theory, that G i Quot(R)H = Quot(R)G [y] = ⊕k−1 i=0 Quot(R) y .

We will will show, using induction, that for every non-negative integer m, R ∩ H

m  i=0

Quot(R) y = R ∩ G i

H

m 

RG y i .

i=0

H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, 185 Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 11, © Springer-Verlag Berlin Heidelberg 2011

186

11 Rings of Invariants which are Hypersurfaces

For m = 0, we need to show that RH ∩ Quot(R)G = RH ∩ RG , i.e., that RH ∩ Quot(R)G = RG . Clearly, RG ⊆ RH ∩ Quot(R)G . Conversely, if h ∈ RH ∩ Quot(R)G , then h is an element of R which is fixed by G and thus h ∈ RG . For the general case, m ≥ 1, we choose a set of left coset representatives, denoted G/H, for H in G. Consider the polynomial  (t − g(y)) = tk + fk−1 tk−1 + · · · + f0 ∈ RG [t] . F (t) = g∈G/H

One of the roots of F (t) is the element y since one of the coset representatives k−1 lies in H. Therefore, y k ∈ i=0 RG y i . m First, we consider the cases m < k. Take f ∈ RH ∩ i=0 Quot(R)G y i . m Write f = i=0 fi y i where fi ∈ Quot(R)G . Since (σ − 1)f lies in the principal ideal of R generated by x, we may write (σ − 1)f = xf  where f  ∈ R. Thus xf  = (σ − 1)f = (σ − 1)

m 

fi y i =

i=0

=

m 

fi (y + x)i −

i=0

= mfm y

m−1

x+

m 

σ(fi y i ) −

i=0

m 

fi y i

i=0

fi y i

i=0 m−2  j=0

m 

 m−1 m−1   m fm y j xm−j + fi (y + x)i − fi y i . j i=0

i=0

 Thus xf  = mfm y m−1 x+w where w ∈ m−2 Quot(R)G y i . Since x ∈ RG , this i=0  m−1 shows that f  = mfm y m−1 + w/x lies in i=0 Quot(R)G y i . Therefore, f  ∈ m−1 Quot(R)G [y]∩R = Quot(R)H ∩R = RH . Hence f  ∈ RH ∩ i=0 Quot(R)G y i m−1 and thus by the induction hypothesis, f  ∈ i=0 RG y i . Since G i Quot(R)H = ⊕k−1 i=0 Quot(R) y and since m < k, we see that {1, y, y 2 , . . . , y m−1 } is linearly independent over Quot(R)G and thus m−1  G i RG y i = ⊕m−1 i=0 R y . i=0

Therefore, mfm y = f  − w/x ∈ RG y m−1 and thus mfm ∈ RG . But m < k ≤ p implies that m is invertible in R and thus fm ∈ RG . Now the induction hypothesis applied to the element m−1

f − fm y m =

m−1  i=0

show that it lies in

fi y i

11 Rings of Invariants which are Hypersurfaces

RH ∩

m−1 

187

RG y i

i=0

m and therefore, f ∈ i=0 RG y i , as required. Thus we have shown that RH ∩ Quot(R)G [y] = RH ∩ RG [y]. Combining this with Quot(R)G [y] = Quot(R)H gives RH ∩ Quot(R)H = RH ∩ RG [y], i.e., RH = RG [y]. This completes the cases m < k. k−1 For the cases m ≥ k, we utilize the fact y k ∈ i=0 RG y i which we showed  m m k−1 above. Using this we have i=0 RG y i = i=0 RG y i and i=0 Quot(R)G y i = k−1 G i i=0 Quot(R) y . Thus for m ≥ k, using the induction hypothesis, we are done. 

The prototypical example of Proposition 11.0.1 is the action of Cp on 2 V2 given in §1.12. Proposition 11.0.2. The hypotheses of Proposition 11.0.1 imply that H is a normal subgroup of index p in G. Proof. Define z := y p − xp−1 y. It is easy to verify that z is fixed by σ and thus z ∈ RG . Consider the polynomial F (t) := tp − xp−1 t − z ∈ RG [t]. Note that for every k ∈ Fp , we have F (y + kx) = (y + kx)p − xp−1 (y + kx) − z = y p + k p xp − xp−1 y − kxp − z = y p − xp−1 y − z + kxp − kxp = 0. Thus p−1 F (t) = k=0 (t − (y + kx)). Since F (t) is G-invariant (having declared that G fixes t), we see that each element g ∈ G must permute the factors of F and hence also the roots of F . Thus for each g ∈ G, there exists k ∈ Fp such that g(y) = y + kx. We define φ : G → Fp by φ(g) = (g(y) − y)/x. Note that for g1 , g2 ∈ G, if g1 (y) = y + k1 x and g2 (y) = y + k2 x, then (g2 g1 )(y) = g2 (g1 (y)) = g2 (y + k1 x) = g2 (y) + k1 x = y + k2 x + k1 x. Thus φ(g2 g1 ) = φ(g2 ) + φ(g1 ) and thus φ is a group homomorphism. Since y ∈ / RG , we know that φ(σ) = 0 and thus φ is a surjective group homomorphism. Furthermore, since y ∈ RH , the subgroup H lies in the kernel of φ. But since H is a maximal proper subgroup of G, this means H is the kernel of φ. Therefore, H is a normal subgroup. Finally, since the index of the kernel of φ equals the cardinality of the image of φ, we see that the index of 

H is |Fp | = p. Example 11.0.3. This example illustrates the use of Proposition 11.0.1. It is also interesting as a simple example of a reflection group whose ring of invariants fails to be a polynomial ring. Let F denote a field of characteristic p containing an element  satisfying p −  = 1. For example, if p −  = 1, we could take F = Fp [], a field of order pp . We consider the representation generated by the three reflections ⎛ ⎛ ⎛ ⎞ ⎞ ⎞ 1000 1000 1000 ⎜a 1 0 0⎟ −1 ⎜0 1 0 0⎟ −1 ⎜0 1 0 0⎟ ⎜ ⎜ ⎟ ⎟ ⎟ α−1 = ⎜ ⎝0 0 1 0⎠ , β = ⎝0 0 1 0⎠ , γ = ⎝0 0 1 0⎠ 0001 0011 101

188

11 Rings of Invariants which are Hypersurfaces

with respect to the standard basis of V . We let {y2 , x2 , y1 , x1 } denote the dual basis of V ∗ . We denote by H the group generated by {α, β, γ} and define G to be the group generated by {α, β, γ, σ} where ⎛ ⎞ 1000 ⎜0 1 0 0⎟ ⎟ σ −1 = ⎜ ⎝0 0 1 0⎠ . 1001 Then G is a Nakajima group and we will use Proposition 11.0.1 to show that H has a hypersurface as a ring of invariants. Further, we let K denote the group generated by {α, β}. Then K ∼ = Cp ×Cp p p−1 p p−1 K and as in §1.12, we have F[V ] = F[x1 , y1 − x1 y1 , x2 , y2 − x2 y2 ]. We define Y1 := y1p − x1p−1 y1 , X1 := (γ − 1)Y1 = (p − )xp1 = xp1 , Y2 := p y2 −x2p−1 y2 and X2 := (γ−1)Y2 = xp1 −x2p−1 x1 . Then F[V ]K = F[Y2 , x2 , Y1 , x1 ]. We will compute F[V ]H = (F[V ]K )H/K by applying Proposition 11.0.1 to R = F[V ]K and the index p subgroup H/K of G/K. Since σ fixes x1 , y1 and x2 and σ(y2 ) = y2 + x1 , we see that σ fixes X1 , Y1 and X2 and σ(Y2 ) = Y2 +(xp1 −x2p−1 x1 ) = Y2 +X2 . Therefore, applying (σ −1) to an arbitrary monomial m = xa1 1 Y1b1 xa2 2 Y2b2 of R gives (σ1 − 1)m = xa1 1 xa2 2 Y1b1 (σ1 − 1)Y2b2 = xa1 1 xa2 2 y2b2 (Y2 + X2 )b2 − Y2b2 b2    b2 = xa1 1 xa2 2 y2b2 Y2b2 −j X2j . j j=1 Thus (σ1 − 1)m is always divisible by X2 and therefore X2 divides (σ1 − 1)f for all f ∈ R. Defining τ by γ = τ σ gives ⎛ ⎞ 1000 ⎜0 1 0 0⎟ ⎟ γ −1 = ⎜ ⎝0 0 1 0⎠ . 001 Then τ fixes y2 , x2 and x1 and τ (y1 ) = y1 + x1 . Therefore, τ fixes Y2 , X2 and X1 and τ (Y1 ) = Y1 + (p − )xp1 = Y1 + xp1 = Y1 + X1 . Working with τ −1 = τ p−1 we find as above that (τ −1 − 1)m is divisible by X1 and thus X1 divides (τ −1 − 1)f for all f ∈ R. Let f ∈ RH be arbitrary. Then τ σ(f ) = f . Thus σ(f ) = τ −1 (f ) and therefore (σ − 1)f = (τ −1 − 1)f . Hence X2 divides (σ − 1)f in R and also X1 divides (σ − 1)f = (τ −1 − 1)f in R. Since R is a unique factorization domain, this implies that the least common multiple X1 X2 /x1 =

11 Rings of Invariants which are Hypersurfaces

189

xp1 (xp−1 − xp−1 ) of X1 and X2 divides (σ − 1)f in R. Thus all the hy1 2 potheses of Proposition 11.0.1 are satisfied and we may apply the Proposition and conclude that RH/K = RG/K [h] where h is any element of R such that (σ − 1)h = xp1 (xp−1 − xp−1 ). For example, we may take h = 1 2 p−1 p−1 y1 ) − xp−1 (y2p − xp−1 y2 ). Thus (X2 Y1 − X1 Y2 )/x1 = (x1 − x2 )(y1p − xp−1 1 1 2 F[V ]H = (F[V ]K )H/K = RG/K [h] = (F[V ]K )G/K [h] = F[V ]G [h]. Hence we have reduced to computing F[V ]G . We may easily find the ring of G-invariants since G is a Nakajima group. We find that F[V ]G = F[x1 , x2 , f1 , f2 ] where f1 = N(y1 ) = Y1p − 2 2 2 2 X1p−1 Y1 = y1p − xp1 −p y p − xp1 −p y1p + x1p −1 y1 , and f2 = N(y2 ) = Y2p − 2 2 X2p−1 Y2 = y2p −x2p −p y2p −(xp1 −x1 xp−1 )p−1 (y2p −xp−1 y2 ). Therefore, F[V ]H = 2 2 p−1 p p−1 p 2p−2 p xp−1 y1 − F[x1 , x2 , f1 , f2 , h] where h = x1 y1 − x2 y1 − x1 y1 + xp−1 1 2 p−1 p p−1 p−1 x1 y2 + x1 x2 y2 . Furthermore, from the Cp representation 2 V2 we know that U p − X2p N1 + p X1 N2 − (X1 X2 )p−1 U = 0 where U = X2 Y1 − X1 Y2 = hx1 , N1 = f1 and N2 = f2 . Therefore, 0 = U p − X2p f1 + X1p f2 − (X1 X2 )p−1 U )p f1 + (xp1 )p f2 − (xp1 )p−1 (xp1 − x1 xp−1 )p−1 (hx1 ) = (hx1 )p − (xp1 − x1 xp−1 2 2 2

− xp−1 )p f1 + x1p = hp − (xp−1 1 2

−p

2

f2 − x1p

−p

(xp−1 − xp−1 )p−1 h . 1 2

12 Separating Invariants

The original and possibly most important use of invariants is to detect whether two mathematical objects are equivalent under some transformation. For example, given two matrices, we may wish to decide whether or not they are conjugate. If their eigenvalues differ, then they cannot be conjugate. In other words, the eigenvalues serve to partially distinguish non-conjugate matrices. Separating invariants typically play a similar role. We note that much of the material in this chapter is valid in characteristic 0. This topic has generated much recent interest, see the book of Derksen and Kemper, [26, §2.3.2, p. 54 and §3.9.4, p. 119], and thesis and papers of Dufresne, [30] and [31], the papers of Kemper, [66], and Draisma, Kemper and Wehlau, [29]. Definition 12.0.1. Let S ⊂ K[V ] and let u, v ∈ V . We will say that u and v are S-equivalent if f (u) = f (v) for all f ∈ S. It is clear that S-equivalence is an equivalence relation on V and also (via restriction) on V . A trivial example is obtained by taking S = K[V ] in which case the equivalence classes in V are the individual points of V . At the opposite end of the spectrum, if we take S = K ⊂ K[V ]G , then all of V is one S-equivalence class. Remark 12.0.2. Note that if S is any subset of K[V ] and R = K[S] is the subalgebra of K[V ] generated by S, then S-equivalence and R-equivalence are the same equivalence relation. For our purposes, the most important example is S = K[V ]G . Theorem 12.0.3. Let G be a finite subgroup of GL(V ) where V is an n dimensional vector space over the field K. Then u, v ∈ V are K[V ]G -equivalent if and only if u ∈ G · v. Proof. Of course, if f (u) = f (v) for any f ∈ K[V ]G , then u and v cannot lie in the same G orbit (since if v = σ · u then f (v) = f (σu) = (σ −1 (f ))(u) = f (u)). H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, 191 Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 12, © Springer-Verlag Berlin Heidelberg 2011

192

12 Separating Invariants

Conversely, suppose u, v ∈ V are K[V ]G -equivalent. Assume, by way of contradiction, that u ∈ / Gv. Then by Corollary 2.1.3, there exists f ∈ K[V ] such that f (u) = 0 and f (σv) = 1 for all σv ∈ Gv. Define h:= NG (f ) ∈ K[V ]G . Since f divides h, we have h(u) = 0. Thus 0 = h(v) = σ∈G (σf )(v). Hence there exists σ ∈ G with 0 = (σf )(v) = f (σ −1 v) contradicting the definition of f .   Note that the hypothesis that G be finite is required in the above theorem. For example, if we consider the subgroup C∗ = GL(1, C) acting on the line ∗ ∗ V = C, then it is clear that C[V ]C = C. Thus there is a single C[V ]C equivalence class. However, there are two orbits in V : one consisting of the origin and a second consisting of all other points of V . Remark 12.0.4. If G is a reductive group over an algebraically closed field K, then it can be shown that u and v are K[V ]G -equivalent if and only if the topological closures of their orbits G · u and G · v in V have non-empty intersection. Theorem 12.0.3 shows that we may detect whether two points v, w ∈ V lie in the same G orbit by evaluating invariants on the two points. Indeed, by Remark 12.0.2, if f1 , f2 , . . . , fr is a generating set for K[V ]G then u and v lie in the same G orbit if and only if fi (u) = fi (v) for all i = 1, 2, . . . , r. Suppose now that f1 , f2 , . . . , fr is a minimal generating set for K[V ]G . The question arises whether it is necessary to verify all r of the equations fi (u) = fi (v) for i = 1, 2, . . . , r in order to be certain that v ∈ G · u. The answer to this question is almost always no. Example 12.0.5. Let K be any field and suppose that the integer n ≥ 3 is invertible in K. Let ξ be a primitive nth root of unity  in K and let G denote ξ 0 the subgroup of GL(V ) generated by σ −1 = . Clearly, G is the cyclic 0ξ ∗ group of order n. If {x, y} is the basis of V dual to the standard basis of V then σ(x) = ξ −1 x and σ(y) = ξ −1 y. It is easy to see that K[V ]G is minimally generated by the n+1 monomials xn , xn−1 y, . . . xy n−1 , y n . However, the values of the three monomials xn , xn−1 y, y n alone suffice to determine whether two points lie in the same orbit. To see this, consider any two points u = (u1 , u2 ) and v = (v1 , v2 ) ∈ V . Suppose that xn (u) = xn (v), xn−1 y(u) = xn−1 y(v) and y n (u) = y n (v). We consider the two possibilities: xn (u) = 0 and xn (u) = 0. If xn (u) = 0, then u lies on the x-axis which consists of two orbits: the origin and the set {(a, 0) | a = 0}. The value y n (u) = un2 = y n (v) distinguishes in which of these two orbits the points u and v lie. Thus we have seen that if xn (u) = 0, then the pair of invariant monomials xn and y n suffice to determine whether u and v lie in the same G-orbit. Next, consider the case where xn (u) = 0. In this case, u1 = 0 and thus

12 Separating Invariants

193

(un−1 u2 )i 1 (un1 )n−i ((xn−1 y)(v))i ((xn−1 y)(u))i = = xi y n−i (v) = n x (u) xn (v)

xi y n−i (u) = ui1 un−i = 2

for all i = 2, 3, . . . , n − 1. Therefore, by Theorem 12.0.3, u and v lie in the same G-orbit. Definition 12.0.6. Let G be a subgroup of GL(V ) where V is an n dimensional vector space over the field. We say that a subset S ⊆ K[V ]G separates if S-equivalence and K[V ]G -equivalence are the same equivalence relation on the points of V . It is easily seen that the following is an equivalent definition. Definition 12.0.7. A subset S ⊂ K[V ]G separates if for all u, v ∈ V , we have the following: if there exists an invariant f ∈ K[V ]G with f (u) = f (v), then there exists h ∈ S with h(u) = h(v). Theorem 12.0.8. Let G be an arbitrary (possibly infinite) group of automorphisms of K[V ]. There is a finite separating set S ⊂ K[V ]G . Proof. We consider the ideal J := (f ⊗ 1 − 1 ⊗ f | f ∈ K[V ]G ) ⊂ K[V ] ⊗K K[V ]. Since the ring K[V ] ⊗K K[V ] ∼ = K[V ⊕ V ] is Noetherian, every generating set for J contains a finite generating set. Thus there exist f1 , f2 , . . . , fm ∈ K[V ]G such that J = (f1 ⊗ 1 − 1 ⊗ f1 , f2 ⊗ 1 − 1 ⊗ f2 , . . . , fm ⊗ 1 − 1 ⊗ fm ). We will show that S = {f1 , f2 , . . . , fm } is a separating set. Let u, v ∈ V G and m suppose there exists f ∈ K[V ] with f (u) = f (v). Write f ⊗ 1 − 1 ⊗ f = i=1 gi (fi ⊗ 1 − 1 ⊗ fi ) with gi ∈ K[V ] ⊗K K[V ]. Define an algebra homomorphism π : K[V ] ⊗K K[V ] −→ K g ⊗ h → g(u)h(v) . Then 0 = f (u) − f (v) = π(f ⊗ 1 − 1 ⊗ f )  m  gi (fi ⊗ 1 − 1 ⊗ fi ) =π i=1

=

m 

π(gi )(fi (u) − fi (v)) .

i=1

Therefore, there exists i such that fi (u) = fi (v).

 

194

12 Separating Invariants

Example 12.0.9. Consider the action of Cp = σ on V = m V2 . As usual, we choose a triangular basis {x1 , y1 , x2 , y2 , . . . , xm , ym } for V ∗ with σ(yi ) = yi +xi and σ(xi ) = xi for all i = 1, 2, . . . , m. Generators for this ring of invariants are given in Theorem 7.4.1. Here we will show that S is a separating subalgebra where S is the subalgebra generated by the following invariants: 1. xi for i = 1, 2, . . . , m. 2. NCp (yi ) = yip − xip−1 yi for i = 1, 2, . . . , m. 3. uij = xj yi − xi yj for 1 ≤ i < j ≤ m. Suppose then that v, w ∈ V and that f (v) = f (w) for all f ∈ S. Write xi (v) = ai , yi (v) = bi , xi (w) = αi and yi (w) = βi for i = 1, 2, . . . , m. First we consider the case where xi (v) = xi (w) = 0 for all i = 1, 2, . . . , m. Then Ni (v) = Ni (w) implies that bpi = βip for all i. Thus 0 = bpi −βip = (bi −βi )p and therefore yi (v) = yi (w) for all i = 1, 2, . . . , m. Thus v = w. Thus we may assume that xk (v) = xk (w) = 0 for some k with 1 ≤ k ≤ m. Let W denote the kth copy of V2 (with basis dual to {xk , yk }). Since F[W ]Cp = F[xk , NCp (yk )], we see that the two invariants xk and NCp (yk ) separate Cp orbits on W . In particular, since ak = αk and bk = βk , there must exist  with 0 ≤  ≤ p − 1 such that (ak , bk ) = σ  · (αk , βk ) = (αk , βk − αk ). Now ujk (v) = ujk (w) for all j = k (where for ease of notation, we define ujk := −ukj if j > k). Thus aj bk −ak bj = αj βk −αk βj . But using bk = βk −αk we see that aj (βk − αk ) − ak bj = αj βk − αk βj . Thus αj αk + αk bj = αk βj . Therefore, αk (bj − βj + αj ) = 0. Since αk = 0 this yields bj = βj − αj for all j = k. Therefore, v = σ  (w) ∈ Cp w. If G is finite, we can give a constructive proof of Theorem 12.0.8. Suppose G is a finite group of automorphisms of K[V ]. Consider the polynomial F (t, z) :=

σ∈G

(t −

n 

σ(xi )z i ) ∈ K[V ]G [t, z]

i=1

Theorem 12.0.10. The coefficients of F (t, z) form a finite separating set. Proof. Suppose that each coefficient of F (t, z) takes the same value at u as it does at v. Then F (t, z)(u) = F (t, z)(v) in the unique factorization domain K[t, z]. Since t − ni=1 xi (u)z i is a factor of F (t, z)(u), it must also be a factor n i of  F (t, z)(v). Therefore, there exists τ ∈ G such that t − i=1 xi (u)z = n n n i i i t− i=1 (τ ·xi )(v)z . Hence i=1 xi (u)z = i=1 (τ ·xi )(v)z and thus xi (u) =   xi (τ −1 · v) for all i = 1, 2, . . . , n. Hence u = τ −1 (v) ∈ Gv. Corollary 12.0.11. If G is a finite subgroup of GL(V ) then there exists a finite separating set S = {f1 , f2 , . . . , fm } with deg(fi ) ≤ |G| for all i = 1, 2, . . . , m. Proof. The coefficients of F (t, z) separate and have degree at most |G|.

 

12.1 Relation Between K[V ]G and Separating Subalgebras

195

12.1 Relation Between K[V ]G and Separating Subalgebras In this section, we prove a number of results that show that each separating subalgebra is closely related to the full ring of invariants. In [87, Lemma 3.4.2], the following result is proved for G any geometrically reductive group. Here we suppose G is finite. Lemma 12.1.1. Let G be a finite group and let R be a K-algebra equipped with a G-action. Suppose f1 , f2 , . . . , fs ∈ RG and consider the ideal I = (f1 , f2 , . . . , fs ) of RG . If f ∈ (f1 , f2 , . . . , fs )R∩RG , then there exists a positive integer t such that f t ∈ I. Proof. The proof is by induction on s. If s = 1, then we may write f = f1 h for some h ∈ R. Since both f1 , f ∈ RG , it follows that f = f1 σ(h) for all σ ∈ G.   |G| Thus σ∈G f = σ∈G f1 σ(h) and therefore f |G| = f1 N(h) ∈ f1 RG . Now assume s ≥ 2 and that the result is true when I has s − 1 generators. Write R = R/fs R and for h ∈ R, denote by h the image of h G in R. Take f ∈ (f1 , f2 , . . . , fs )R ∩ RG . Then f ∈ (f1 , f2 , . . . , fs−1 )R ∩ R . Thus by our induction hypothesis, there exists a positive integer t such that s t G f ∈ I = (f1 , f2 , . . . , fs−1 )R . Hence we may write f t = i=1 fi hi where s−1 G t h1 , h2 , . . . , hs ∈ R and h1 , h2 , . . . , hs−1 ∈ R . Therefore f − i=1 fi hi =   s−1 fs hs and f t − i=1 fi σ(hi ) = fs σ(hs ) for all σ ∈ G. Hence σ∈G (f t − s−1  |G| t|G| −fs N(hs ) ∈ (f1 , f2 , . . . , fs−1 )R∩ σ∈G fs σ(hs ). Thus f i=1 fi σ(hi )) = G R . Applying the induction hypothesis again, there is a positive integer v |G| such that (f t|G| − fs N(hs ))v ∈ (f1 , f2 , . . . , fs−1 )RG . Therefore f t|G|v ∈ G (f1 , f2 , . . . , fs )R .   Theorem 12.1.2. Suppose that S ⊂ K[V ]G is a graded separating subalgebra. Then K[V ]G is integral over S. Proof. Let I = S+ · K[V ] denote the ideal of K[V ] generated by S+ and let J = K[V ]G + · K[V ] denote the Hilbert ideal. Take any v ∈ V(I). Since S is a separating set and since h(v) = 0 for h ∈ S+ , it follows that f (v) √ = 0√for all f ∈ K[V ]G ∈ V(J). Hence V(I) ⊆ V(J). Therefore J ⊆ I. + . Thus v √ I ∩ K[V ]G . Take f ∈ K[V ]G In particular, K[V ]G + ⊆ + . Then, by the above, √ f ∈ I. Therefore, there exists a positive integer m such that f m ∈ I ∩K[V ]G . By Lemma 12.1, there is a positive integer t with f mt ∈ S+ K[V ]G . This implies that the ring K[V ]G /S+ K[V ]G has Krull dimension 0 and thus this ring is a finite dimensional K vector space. Therefore by the graded Nakayama lemma 2.10.1, we see that K[V ]G is a finitely generated S-module. This implies that K[V ]G is integral over S.   In particular, we have the following result.

196

12 Separating Invariants

Corollary 12.1.3. If S ⊂ K[V ]G is a graded separating subalgebra, then dim S = dim K[V ]G . Remark 12.1.4. It is possible to show (see [30][Corollary 3.2.5]) that the hypothesis that S be graded is not required for the proof of Corollary 12.1.3 provided that G is reductive. Note that the converse of Theorem 12.1.2 is not true as the following example shows. Example 12.1.5. Consider the two dimensional representation of the trivial group, G = {e} over a field K. Take S = K[x, y 2 , x2 y] ⊂ K[V ]G = K[x, y]. Since y = (x2 y)/(x)2 ∈ Quot(S) and since y satisfies the monic equation T 2 − y 2 ∈ S[T ], we see that y lies in the integral closure of S. Hence K[x, y] = K[V ]G is the integral closure of S. Consider the two points u = (0, 1) and v = (0, −1). Since x(u) = 0 = x(v) and y 2 (u) = 1 = y 2 (v) and x2 y(u) = 0 = x2 y(v) we see that S does not separate u from v and thus S is not a separating subalgebra. Theorem 12.1.6. Suppose G is finite and S ⊂ K[V ]G is a graded separating subalgebra. Then the field Quot(K[V ]G ) is a finite purely inseparable extension of Quot(S). Proof. Dual to the inclusion φ : S → K[V ]G is the morphism φ∗ : V //G → Spec(S). In terms of ideals, this map is given by φ∗ (P ) = P ∩ S for a prime ideal P ∈ V //G = Spec(K[V ]G ). Since φ is injective, φ∗ is dominant, i.e., its image is dense in Spec(S). We claim that φ∗ is also injective. First, we show it is injective on maximal ideals. Take two different maximal ideals I and J of K[V ]G . The ideals I and J have height n = dim V . Hence both V V (I) and V V (J) are zero dimensional subvarities of V , i.e., both these varieties consist of a (non-empty) finite set of points of V . Write V V (I) = there exists {a1 , a2 , . . . , as } and V V (J) = {b1 , b2 , . . . , bt }. Since S separates,  fi ∈ S with fi (ai ) = fi (b1 ) for all i = 1, 2, . . . , s. Define f := si=1 (fi −fi (ai )). Then f ∈ S with f (ai ) = 0 for all i and f (b1 ) = 0. Therefore, f ∈ S∩I = φ∗ (I) but f ∈ / S ∩ J = φ∗ (J) and φ∗ (I) = φ∗ (J). Now we use this to show that φ∗ is injective. Take any two different prime ideals P and Q of K[V ]G . Assume, by way of contradiction, that φ∗ (P ) = φ∗ (Q). Since the prime ideals P and Q are different, there exists (without loss of generality) a maximal ideal I of K[V ]G containing P but not containing Q (see for example [80][Theorem 5.5]). Since φ∗ (P ) = φ∗ (Q) and P ⊂ I we have the chain of prime ideals φ∗ (Q) ⊂ ∗ φ (I) in S. By Theorem 12.1.2, the “Going Up” Theorem (Theorem 2.5.2 (2)) applies here. Thus there exists a prime ideal J of K[V ]G such that Q ⊂ J and J ∩ S = φ∗ (I). By “Going Up” and “Going Down” (Theorem 2.5.2), since I is maximal in K[V ]G , the ideal I ∩ S = J ∩ S is maximal in S and this implies that J is maximal in K[V ]G . But then φ∗ (I) = φ∗ (J) implies that I = J since φ∗ is injective on maximal ideals. However, Q ⊂ J and Q ⊆ I. This contradiction shows that φ∗ is indeed injective.

12.1 Relation Between K[V ]G and Separating Subalgebras

197

By [56, Theorem 4.6], since φ∗ is a dominant injective morphism, we have that Quot(K[V ]G ) is a finite purely inseparable extension of φ(Quot(S)) hence also of Quot(S).   Remark 12.1.7. The above theorem is true even without the hypothesis that S be graded. For a proof of this see [26, Proposition 2.3.10] or [30]. Definition 12.1.8. Let S be a subalgebra of K[V ]. If K has characteristic p then we define the inseparable closure or p-root closure of S in K[V ], denoted S by t S := {f ∈ K[V ] | ∃ t ∈ N with f p ∈ S}. For ease of notation, we define S = S when K has characteristic 0. Proposition 12.1.9. Let S be a subalgebra of K[V ]G and suppose that S separates. Then S also separates. Proof. If K has characteristic 0 then there is nothing to prove. Thus we suppose that the characteristic of K is p > 0. Take u, v ∈ V and suppose that r r r h(u) = h(v) for all h ∈ S. Then 0 = (h(u) − h(v))p = hp (u) − hp (v) for

Since S separates h ∈ S and all r ∈ N. Therefore f (u) = f (v) for all f ∈ S.   this implies that u and v are K[V ]G -equivalent. Thus S separates. Theorem 12.1.10. Let S be a graded separating subalgebra of K[V ]G . Then

S = K[V ]G .

Proof. Since K[V ]G is integrally closed, we know that S ⊆ K[V ]G . For the r opposite inclusion, take f ∈ F[V ]G . Then Theorem 12.1.6 implies that f p ∈ r Quot(S) for some r ∈ N (use r = 0 and f p = f if K has characteristic 0). r r  Furthermore, by Theorem 12.1.2, f p is integral over S, and thus f p ∈ S.

 Thus f ∈ S.  



Proposition 12.1.11. Let S be a subalgebra of K[V ]. Then S = S. 

and f satisfies some equation z t + Proof. Take f ∈ S. Then f ∈ Quot(S) t−1 + · · · + c0 = 0 where c0 , c1 , . . . , ct ∈ S. We must show that there ct−1 z r  Write f = a/b where a, b ∈ S.

Then there exists exists r ∈ N with f p ∈ S.  r r p pm ∈ S. Take r ≥ max{, m}. Then ap , bp ∈ S and , m ∈ N with a , b r r r r f p = ap /bp ∈ Quot(S). Furthermore, 0 = (f t + ct−1 f t−1 + · · · + c0 )p = r r r r r r r r (f t )p + cpt−1 (f t−1 )p + · · · + cp0 = (f p )t + cpt−1 (f p )t−1 + · · · + cp0 with r r r r

 cp0 , cp1 , . . . , cpt−1 ∈ S. Thus f p ∈ S which means that f ∈ S.

 For the opposite inclusion, take f ∈ S. Then there exists r ∈ N with r  Write h = a/b with a, b ∈ S. Then bf pr = a ∈ S. Thus (bf )pr = h = f p ∈ S. r r r

Thus f = bf /b ∈ Quot(S).

bp −1 (bf p ) = bp −1 a ∈ S and hence bf ∈ S.

198

12 Separating Invariants

Furthermore, if h satisfies the monic equation z t + ct−1 z t−1 + · · · + c0 = 0 with

then f satisfies the monic equation z tpr +ct−1 z (t−1)pr + c0 , c1 , . . . , ct ∈ S ⊂ S, 

· · · + c0 = 0. Therefore, f ∈ S.   Remark 12.1.12. In fact, it can be shown that if S is a graded subalgebra of F[V ]G where F is a field of characteristic p, then S separates if and only if S = F[V ]G . For details see [50] or [27, Remark 1.3].

12.2 Polynomial Separating Algebras and Serre’s Theorem The ideal J := (f ⊗ 1 − 1 ⊗ f | f ∈ K[V ]G ) ⊂ K[V ] ⊗K K[V ] plays a central role in the proof of Theorem 12.0.8. We consider this ideal more carefully now. This ideal corresponds to the reduced scheme having the same underlying topological space as V ×V //G V . Following Dufresne we call this the separating scheme and denote it by SG . Thus SG = Spec(K[V ] ⊗K[V ]G K[V ]) . We consider the map δ : K[V ] → K[V ] ⊗K K[V ] defined by δ(f ) = f ⊗ 1 − 1 ⊗ f . For any subalgebra B ⊆ K[V ], we have K[V ] ⊗K K[V ] K[V ] ⊗B K[V ] ∼ = (δ(B)) and in particular, K[V ] ⊗K K[V ] K[V ] ⊗K K[V ] K[V ] ⊗K[V ]G K[V ] ∼ = . = G (δ(K[V ] )) J Theorem 12.2.1 ([30] Theorem 3.2.1). Let B be a subalgebra of K[V ]G . Then B is a separating subalgebra if and only if (δ(B)) = (δ(K[V ]G ). Proof. We will exploit the fact that the radical of any ideal I is obtained by taking the intersection of all maximal ideals containing I. Thus we need to consider the maximal ideals of K[V ] ⊗K K[V ]. These are precisely the ideals of the form IV ×V {(u, v)} ∩ (K[V ] ⊗K K[V ]). (Recall that IV ×V {(u, v)} denotes the maximal ideal of K[V ]⊗K K[V ] corresponding to a geometric point (u, v) ∈ V × V .) The assertion that B is a separating subalgebra means that every pair of points u, v ∈ V can be separated by invariants if and only if they can be separated by elements of B. In terms of ideals, this can be stated as IV ×V (u, v) ∩ (K[V ] ⊗K K[V ]) ⊇ δ(K[V ]G ) if and only if

12.2 Polynomial Separating Algebras and Serre’s Theorem

IV ×V (u, v) ∩ (K[V ] ⊗K K[V ]) ⊇ δ(B) for all u, v ∈ V . Thus B is a separating subalgebra if and only if (δ(K[V ]G )).

199

(δ(B)) =  

If G is a finite group, then the separating scheme is quite simple; it is a union of |G| subspaces each of dimension n = dim V lying in V × V . Indeed, take τ ∈ G and let Hτ denote the subspace Hτ := {(v, τ (v)) | v ∈ V } of V ×V. Lemma 12.2.2. If G is a finite group, then

SG = Hτ . τ ∈G

Proof. Since G is finite, the invariants will separate points not in the same G-orbit by Theorem 12.0.3. Thus (u, v) ∈ SG if and only if IV ×V {(u, v)} ∩ (K[V ] ⊗K K[V ]) ⊇ δ(K[V ]G ), if and only if v = τ (u) for some τ ∈ G, if and   only if (u, v) ∈ Hτ for some τ ∈ G. A natural question is whether we may find separating subalgebras which are better behaved than the full ring of invariants. In particular, when may we find a polynomial subalgebra which separates? The following theorem answers this question in characteristic 0 and gives a necessary answer in positive characteristic. Theorem 12.2.3. Let G be a finite group. If K[V ]G contains a graded separating algebra which is a polynomial ring, then the action of G on V is generated by reflections. Proof. Suppose B ⊆ K[V ]G is a graded polynomial separating algebra. By Corollary 12.1.3, the algebra B has dimension n. Write B = K[f1 , f2 , . . . , fn ]. Then (δ(B)) = (δ(f1 ), δ(f2 ), . . . , δ(fn )) has n generators and cuts out a variety SG = ∪τ ∈G Hτ of codimension n in V × V . Thus δ(f1 ), δ(f2 ), . . . , δ(fn ) is a partial homogeneous system of parameters. Therefore   K[V ] ⊗K K[V ] Spec(K[V ] ⊗B K[V ]) ∼ = Spec (δ(B)) is a complete intersection and hence is also Cohen-Macaulay. Therefore, since K[V ] ⊗K K[V ] is Noetherian, Hartshorne’s Connectedness Theorem, see [51][Corollary 2.4], applies and we conclude from it that Spec(K[V ] ⊗B K[V ]) is connected in co-dimension 1. Since Spec(K[V ] ⊗B K[V ]) and SG correspond to the same underlying topological space, this means that SG is connected in co-dimension 1. Consider the two components He and Hσ of SG corresponding to the identity and an arbitrary element σ of G. Since SG is connected in co-dimension 1, there exists a sequence of components He = Hτ0 , Hτ1 , . . . , Hτs = Hσ

200

12 Separating Invariants

such that Hτi−1 ∩ Hτi has dimension n − 1 (i.e., codimension 1 in Hτi−1 and in Hτi ) for all i = 1, 2, . . . , s. Now (u, v) ∈ Hτi−1 ∩ Hτi if and only if −1 τi−1 (u) = v = τi (u), if and only if τi−1 τi (u) = u and v = τi (u), if and only if −1 −1 τi−1 τi u∈V and v = τi (u). Therefore, Hτ ∩ Hτ ∼ = V τi−1 τi . This means that i−1

−1 τi−1 τi

i

−1 τi−1 τi

= dim Hτi−1 ∩ Hτi = n − 1, i.e., is a reflection. Therefore, dim V −1 σ = τs = τ0 (τ0−1 τ1 )(τ1−1 τ2 ) · · · (τs−1 τs ) expresses σ as a product of reflections in G.   Remark 12.2.4. By Remark 12.1.4 we may omit the hypothesis of graded from Theorem 12.2.3. Corollary 12.2.5 (Serre’s Theorem). Let G be a finite group. If K[V ]G is a polynomial ring, then the action of G on V is generated by reflections. In characteristic zero, Theorem 12.1.10 shows that if S is a graded polynomial separating subalgebra, then S = K[V ]G . In positive characteristic, suppose F[V ]G = F[f1 , f2 , . . . , fn ] is a polynomial ring. Then using Proposir1 r2 rn tion 12.1.9, we see that S = F[f1p , f2p , . . . , fnp ] is a polynomial separating subalgebra for all r1 , r2 , . . . , rn ∈ N. These two results together with Theorem 12.2.3 suggest that perhaps whenever there is a polynomial separating subalgebra, the ring of invariants itself is polynomial. The following example, due to Dufresne [30] shows that this is not the case. Example 12.2.6. In this example, we show that for positive characteristic, the existence of a polynomial separating subalgebra, S ⊂ F[V ]G , does not necessarily imply that F[V ]G is a polynomial ring. To see this, we consider Example 11.0.3 from the point of view of separating invariants. We maintain the notation from Example 11.0.3, although we will now denote the group of order p3 by G. In that example, we computed the ring of invariants of a four dimensional representation V of G, the elementary Abelian group of order p3 over a field F. We found that F[V ]G = F[x1 , f1 , x2 , f2 , h] where 2

2

f1 = y1p − xp1

−p p

2

y − xp1

−p p y1

2

+ x1p

−1

y1 ,

f2 = y2p − xp2 −p y2p − (xp1 − x1 x2p−1 )p−1 (y2p − x2p−1 y2 ) and h = (x1p−1 − x2p−1 )(y1p − x1p−1 y1 ) − x1p−1 (y2p − x2p−1 y2 ) 2

2

= x1p−1 y1p − x2p−1 y1p − x12p−2 y1p + x1p−1 x2p−1 y1 − x1p−1 y2p + x1p−1 x2p−1 y2 with 2

hp − (x1p−1 − x2p−1 )p f1 + xp1

−p

2

f2 − xp1

−p

We define f2 , a perturbation of f2 , as follows,

(x1p−1 − x2p−1 )p−1 h = 0 .

12.3 Polarization and Separating Invariants

201

f2 := f2 − (x1p−1 − x2p−1 )p−1 h and take S = F[x1 , x2 , f1 , f2 ] = F[x1 , x2 , f1 , f2 ]. Note that f2 is homogeneous of the same degree as f2 . Then F[V ]G = F[x1 , f1 , x2 , f2 , h] = S[h] with hp − (x1p−1 − x2p−1 )p f1 + (x1p−1 )p f2 = 0 .

This shows that S = F[V ]G and In particular, hp ∈ S and therefore h ∈ S. thus by Proposition 12.1.9, S is a separating subalgebra. Since x1 , x2 , f1 , f2 forms a homogeneous system of parameters, these elements are algebraically independent and thus S is a polynomial subalgebra of F[V ]G which is also a separating algebra.

12.3 Polarization and Separating Invariants Recall the polarization operators defined in §1.9. In this section, we prove a surprising theorem relating separating invariants and polarization. This material comes from the work of Draisma, Kemper and Wehlau [29]. Theorem 12.3.1. Suppose S is a separating subset of K[V ]G and let m ≥ 1. Then Polm (S) is a separating subset of K[m V ]G . More generally, suppose S is a separating subset of K[W1 ⊕ W2 ⊕ · · · ⊕ Wr ]G and that m1 , m2 , . . . , mr are positive integers. Then Polm1 ,m2 ,...,mr (S) is a separating subset of K[m1 W1 ⊕ m2 W2 ⊕ · · · ⊕ mr Wr ]G . Proof. We will prove the second assertion. Consider two points (1)

(1)

(2)

(2)

(r)

(r)

(1) (2) (r) v = (v1 , v2 , . . . , vm , v1 , v2 , . . . , vm , . . . , v1 , v2 , . . . , vm ), 1 2 r (1)

(1)

(2)

(2)

(r)

(r)

(1) (2) (r) , w1 , w2 , . . . , wm , . . . , w1 , w2 , . . . , wm ) w = (w1 , w2 , . . . , wm 1 2 r

of V where V = m1 W1 ⊕ m2 W2 ⊕ · · · ⊕ mr Wr . Suppose that g(v) = g(w) for all g ∈ Polm1 ,m2 ,...,mr (S) = ∪f ∈S Polm1 ,m2 ,...,mr (f ). We need to show that this assumption implies that v and w lie in the same G-orbit. (k) Take f ∈ S and for k = 1, 2, . . . , r and i = 1, 2, . . . , mk , let λi denote an indeterminate. Consider f

m1  i=1

(1) (1)

λi v i

,

m2 

(2) (2)

λi vi

i=1

,...,

mr 

(r) (r)



λi vi

=



i=1

(1)

J

where each fJ lies in Polm1 ,m2 ,...,mr (S). Therefore (1)

(1)

(1)

(1)

(1)

(r) λJ fJ (v1 , v2 , . . . , vm ) r

(r) (r) fJ (v1 , v2 , . . . , vm ) = fJ (w1 , w2 , . . . , wm ) r r

202

12 Separating Invariants

for all J and m1 m2 mr     (1) (1) (2) (2) (r) (r) λi vi , λi vi , . . . , λi vi f i=1

i=1

=f

m1 

(1)

i=1 (1)

λi wi

,

i=1

m2 

(2)

(2)

λi wi

,...,

i=1

mr 

(r)

(r)



λi wi

.

i=1 (k)

Now pick α ∈ K and substitute in the above λi by αi−1 for all k = 1, 2, . . . , r. Thus m1 m2 mr     (1) (2) (r) αi−1 vi , αi−1 vi , . . . , αi−1 vi f i=1

=f

m1 

i=1

i=1 (1)

αi−1 wi

,

i=1

m2 

(2)

αi−1 wi

,...,

mr 

i=1

(r)

αi−1 wi

 .

i=1

Since this holds for all f in the separating set S, there must exist σα ∈ G such that m1 m2 mr     (1) (2) (r) σα αi−1 vi , αi−1 vi , . . . , αi−1 vi i=1

=

m1  i=1

i=1

i=1 (1)

αi−1 wi

,

m2 

(2)

αi−1 wi

i=1

,...,

mr 

(r)

αi−1 wi

 .

i=1

Now to each α ∈ K, we have associated an element σα of G. Let t := max{m1 , m2 , . . . , mr }. Choose τ ∈ G such that there exist α1 , α2 , . . . , αt distinct elements of Kwith σαs = τ for all s = 1, 2, . . . , t.  mk i−1 (k) mk i−1 (k)  αs wi Thus we have τ = ( i=1 for all 1 ≤ s ≤ t and i=1 αs vi 1 ≤ k ≤ r. Writing this in matrix terms we have ⎞ ⎛ ⎞ ⎞⎛ ⎛ (k) (k) τ (v1 ) − w1 0 1 α1 α12 . . . α1mk −1 (k) (k) ⎟ ⎜0⎟ ⎜1 α2 α2 . . . αmk −1 ⎟ ⎜ ⎟ ⎜ τ (v ) − w 2 2 2 2 ⎟ ⎜ ⎟ ⎟ ⎜ ⎜. . .⎟ .. .. ⎟ ⎜ .. ⎟=⎜ ⎜ ⎠ ⎝ .. .. . . ⎠ ⎝ .. ⎠ ⎝ . 2 mk −1 (k) (k) 0 . . . αm 1 αmk αm τ (vm ) − wm k k k

k

for all 1 ≤ k ≤ r. Since the square matrix is a Vandermonde matrix with  (k) (k) determinant 1≤i<j≤j (αj − αi ) = 0, we see that τ (vi ) − wi = 0 for all 1 ≤ i ≤ mk and 1 ≤ k ≤ r. Thus τ (v) = w.   Example 12.3.2. We consider the representation 3 V2 of C3 over a field F of characteristic 3. By Theorem 1.11.2, we know that F[V2 ]C3 = F[N, x] where V2∗ has triangular basis {x, y} and N = N(y) = y 3 −x2 y. Let {x1 , y1 , x2 , y2 , x3 , y3 } be a triangular basis for (3 V2 )∗ and consider the Pol3 ({x, N }). Clearly, Pol(x) = {x1 , x2 , x3 }. To compute Pol(N ), we consider

12.3 Polarization and Separating Invariants

203

(λ1 y1 + λ2 y2 + λ3 y3 )3 − (λ1 x1 + λ2 x2 + λ3 x3 )2 (λ1 y1 + λ2 y2 + λ3 y3 ) = λ31 N1 + λ32 N2 + λ33 N3 + λ21 λ2 x1 u12 + λ21 λ3 x1 u23 − λ1 λ22 x2 u12 − λ1 λ23 x3 u13 + λ22 λ3 x2 u23 − λ2 λ23 x3 u23 − λ1 λ2 λ3 (x2 u13 + x1 u23 ) where Ni = N(yi ) and uij = xi yj − xj yi for 1 ≤ i < j ≤ 3. Thus Pol(N ) = {N1 , N2 , N3 } ∪ {xk uij | k = 1, 2, 3, 1 ≤ i < j ≤ 3}. Therefore by Theorem 12.3.1, we see that the set A = {x1 , x2 , x3 , N1 , N2 , N3 } ∪ {xk uij | k = 1, 2, 3, 1 ≤ i < j ≤ 3} is a separating set for the action of C3 on 3 V2 . Note that u312 = x32 N1 − x31 N2 + (x1 x2 )2 u12 and thus we see that u12 ∈ S where S is the F-algebra generated by the separating set A. Similarly,

u13 , u23 ∈ S. Comparing Example 1.13.3 with Theorem 12.1.10 we see that the four other generators of F[3 V2 ]C3 , TrC3 (y12 y22 y3 ), TrC3 (y12 y2 y32 ), TrC3 (y1 y22 y32 ) and TrC3 (y12 y22 y32 )

 also lie in S.

13 Using SAGBI Bases to Compute Rings of Invariants

In [98], Shank constructed generating sets, in fact, SAGBI bases, for the rings F[V4 ]Cp and F[V5 ]Cp for all primes p ≥ 5. Of course, for the primes p = 2, 3, the corresponding actions are actions of Cp2 or Cp3 , not Cp . The rings of invariants F[V4 ]C4 , F[V4 ]C9 , F[V5 ]C8 and F[V5 ]C9 are all easily computed by computer, for example, using MAGMA. We explain here the method used in [98]. This is a method used to confirm that a conjectural set of generators does indeed generate the full ring of invariants. In theory, this method will work for any ring of invariants F[V ]G provided we can compute a closed form for its Hilbert series, H(F[V ]G , λ). Currently, for modular groups, there is only a formula for H(F[V ]G , λ) for those groups G for which p2 does not divide |G| (and for permutation representations). This formula is described in [55]. We begin with a simple example to illustrate the main ideas. Example 13.0.1. Let G = Cp , V = V2 ⊕ V2 and let F be any field of characteristic p. We choose a generator σ of Cp and a basis {x1 , y1 , x2 , y2 } for V ∗ such that σ(xi ) = xi and σ(yi ) = yi + xi for i = 1, 2. We want to show that F[V ]G = F[x1 , x2 , u, N1 , N2 ] where u = x2 y1 − x1 y2 , and Ni = yip − xip−1 yi for i = 1, 2. The method of Hughes and Kemper gives H(F[V ]G , λ) =

(1 −

(1 − λ2 )p . − λ)2 (1 − λp )2

λ2 )(1

We declare that x1 < y1 < x2 < y2 and work with the resulting graded reverse lexicographic ordering on the monomials of F[V ] = F[x1 , y1 , x2 , y2 ]. With this ordering we have LM(xi ) = xi , LM(Ni ) = yip for i = 1, 2 and LM(u) = x2 y1 . We define T := F[x1 , N1 , x2 , N2 , u]. It is easily verified that these five functions are all invariants and therefore, T ⊂ F[V ]G . We want to show that T is the full ring of invariants. We consider the monomial algebras Q := F[x1 , y1p , x2 , y2p , x2 y1 ] ⊆ LT(T ) and A := F[x1 , y1p , x2 , y2p ]. Since x1 , y1p , x2 , y2p is H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, 205 Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 13, © Springer-Verlag Berlin Heidelberg 2011

206

13 Using SAGBI Bases to Compute Rings of Invariants

a homogeneous system of parameters for F[V ], we see that Q must be a finite A-module. p−1 A (x2 y1 )i , but rather than It is not too difficult to see that Q = p−1 i=0 prove this directly we consider M := i=0 A (x2 y1 )i as an A-submodule of i Q. It is clear that M is a free A-module: M = ⊕p−1 i=0 A (x2 y1 ) . Thus H(M, λ) =

p−1 

H(A (x2 y1 )i , λ)

i=0



p−1

=

H(A, λ)λ2i

i=0

=

p−1  1 λ2i (1 − λ)2 (1 − λp )2 i=0

= H(F[V ]G , λ). Thus we have the chain of inclusions M ⊆ Q ⊆ LT(T ) ⊆ LT(F[V ]G ). But since H(F[V ]G , λ) = H(LT(F[V ]G ), λ), we have H(M, λ) = H(F[V ]G , λ), and therefore, M = Q = LT(T ) = LT(F[V ]G ). In particular, the monomials in M form a vector space basis for the vector space of lead terms of F[V ]G . Since the monomials in M are generated multiplicatively by x1 , x2 , y1p = LT(N1 ), y2p = LT(N2 ) and x2 y1 = LT(u), we have, in fact, proved that {x1 , x2 , N1 , N2 , u} is a SAGBI basis (and hence a generating set) for F[V ]G . There are some features of Example 13.0.1 that make the computation simpler than it is in general. These include the fact that the natural minimal algebra generating set is already a SAGBI basis and the fact that monomial algebra Q is a free A-module in the example. In particular, this second fact made the calculation of H(M, λ) particularly easy. We will have to account for less well behaved algebras in the general case. Having seen this first example we proceed to describe the method in general. Suppose then that we have a closed expression for H(F[V ]G , λ). The first step is to somehow guess a finite set of invariants B which is a SAGBI basis for F[V ]G . The method then gives a way to use the knowledge of H(F[V ]G , λ) to prove that B is indeed a SAGBI basis for F[V ]G . Consider the algebra T generated by the elements of B and the algebra Q generated by the set of monomials {LM(f ) | f ∈ B}. Clearly Q ⊆ LT(T ). Let n denote the Krull dimension of F[V ]. We choose n invariants f1 , f2 , . . . , fn in F[V ]G such that a1 , a2 , . . . , an forms a homogeneous system of parameters for F[V ] where ai = LM(fi ) for i = 1, 2, . . . , n. Then Q (and T ) is a finitely generated A-module and so there exists a finite set Γ0 ⊂ Q such that  A·h . Q= h∈Γ0

We wish to find such a set Γ0 . We choose a finite subset Γ of Q and consider the A-module, M , generated by Γ :

13 Using SAGBI Bases to Compute Rings of Invariants

M :=



207

A·h .

h∈Γ

Thus we have M ⊆ Q ⊆ LT(T ) ⊆ LT(F[V ]G . If we can show M = LT(F[V ]G ), then we will have shown that B and also Γ ∪ {a1 , a2 , . . . , an } is a SAGBI basis for F[V ]G . Since M ⊆ LT(F[V ]G ), it suffices to show that H(M, λ) = H(LT(F[V ]G ), λ). But since H(LT(F[V ]G ), λ) = H(F[V ]G , λ), our goal is to prove that H(M, λ) = H(F[V ]G , λ). Since we have a closed expression for H(F[V ]G , λ), we need to compute H(M, λ) from knowledge of Γ . This computation is at the heart of this method. We exploit the fact that M is an A-module and A is generated by monomials. We first decompose M into a direct sum of certain A-submodules and then proceed to compute the Hilbert series for each of these submodules. We illustrate the entire method by computing a generating set for F[V4 ]C11 . This example displays essentially all of the details involved in computing F[V4 ]Cp for general p. Example 13.0.2. We write R := F[V4 ]C11 where F is a field of characteristic 11. We take the triangular basis {x, y, z, w} of V4 where σ(x) = x, σ(y) = y + x, σ(z) = z + y and σ(w) = w + z. Again, we use the reverse lexicographic ordering with x < y < z < w. Thus by Theorem 5.2.4, there is indeed a finite SAGBI basis for F[V4 ]C11 . We begin by choosing h1 = x, h2 = d = y 2 − 2xz − xy, h3 = Tr(wp−1 ) = Tr(w10 ) = z 10 + . . . and h4 = N(w) = w11 + . . .. Then the corresponding lead terms a1 = x, a2 = y2 , a3 = z 10 , a4 = w 11 do indeed form a homogeneous system of parameters for F[V ]. There are two more integral invariants: e := y 3 − 3xyz + x2

and

f := 3z 2 y 2 − 6wy 3 − 3zy 3 − y 4 − 8z 3 x + 18wzyx + 6z 2 yx + 9wy 2 x + 9zy 2 x + 2y 3 x − 9w2 x2 − 18wzx2 − 12z 2 x2 − 3wyx2 − 6zyx2 − y 2 x2 . We make the “guess” that the set B comprised of the following twenty-four invariants is a SAGBI basis for R. (0) h1 , h2 , h3 , h4 (1) e, f (2) Tr(z i w10 ) for 1 ≤ i ≤ 9 (3) Tr(z i w5 ) for 3 ≤ i ≤ 9 (4) Tr(wi ) for 7 ≤ i ≤ 9 (5) Tr(z 2 wi ) for 7 ≤ i ≤ 9 In reality, the corresponding guess for general p was motivated by computer experiments with small values of p. Using Lemma 9.0.2 it is not too hard to show that the corresponding lead monomials are

208

13 Using SAGBI Bases to Compute Rings of Invariants

(0) x, y 2 , z 10 , w11 (1) y 3 , y 2 z 2 (2) z 10+i for 1 ≤ i ≤ 9 (3) yz 8+i for 3 ≤ i ≤ 9 (4) y 10−i z 2i−10 for 7 ≤ i ≤ 9 (5) y 10−j z 2j−8 for 7 ≤ j ≤ 9 We take A := F[x, y 2 , z 10 , w11 ] and consider R as an A-module. Now we wish to guess our set Γ of A-module generators for R. We take Γ1 to be the above listed monomials excepting x, y 2 , z 10 , w11 which, of course, lie in A. Thus Γ1 consists of 24 monomials. To form Γ , we add four more monomials: (6) 1 (7) yz 19 = LM(Tr(zw10 ) Tr(w9 )) = LM(Tr(zw 10 )) LM(Tr(w9 )) (8) y 4 z 4 = LM(f 2 ) = LM(f )2 (9) y 5 z 2 = LM(ef ) = LM(e) LM(f ) Note that each of these four extra monomials lies in LT (F[B]). We let M denote the A-module generated by the twenty-eight monomials of Γ . We will show, using their Hilbert series, that M = LT(F[V4 ]C11 ) and thus that Γ ∪ {x, y2 , z 10 , w11 } generate LT(F[V4 ]C11 ) as an algebra. We are faced with the problem of calculating H(M, λ). In Example 13.0.1, this was easily done because M was a free A-module. Here this is definitely not the case. We proceed by decomposing M into a direct sum of simpler A-modules as follows. For a, b ∈ N we define M(a,b) := spanF {xi y j z k w | j ≡ a (mod 2) and k ≡ b (mod 10)} . Then each M(a,b) is an A-module and M = ⊕1a=0 ⊕9b=0 M(a,b) , and therefore, H(M, λ) =

1  9 

H(M(a,b) , λ) .

(13.0.1)

a=0 b=0

Thus we have reduced to finding the Hilbert series of the simpler A-modules, M(a,b) . Writing each of the twenty-eight module generators for M in the form xi y j z k w , we note that i and  are always zero. It is clear that for an Amodule generator, i must be zero since x ∈ A. There are deeper reasons why we always find  = 0. This is related to our choice of w as the greatest variable in our triangular basis. We begin by sorting these twenty-eight A-module generators of M according to the parity of j and the residue class of k modulo 10.

13 Using SAGBI Bases to Compute Rings of Invariants

209

k mod 10 0 0

1

3

y2 z2

1 z 11

j mod 2 1

2

y3

z 12 y5 z2

4

5

y4z4 z 13

6

7

y2z6

z 14

z 15

y3z4

z 16

8

9

y2z8 z 17

y3z6

z 18

z 19

yz 8

yz 10 yz 11 yz 12 yz 13 yz 14 yz 15 yz 16 yz 17

yz 19

Thus M(j,k) is cyclic for j = 0, 1 and k odd. Therefore H(M(j,k) , λ) = H(A, λ)λ10+k+j for j = 0, 1 and k odd. Also H(M(0,0) , λ) = H(A, λ) and H(M(1,8) , λ) = H(A, λ)λ9 . For the other values of (j, k), the A-module M(j,k) has exactly two generators and is not a free A-module. Suppose M(j,k) is minimally generated by the two monomials m1 = y a1 z b1 and m1 = y a2 z b2 . Note that m1 cannot divide m2 nor can m2 divide m1 since both are required as generators of M(j,k) . For concreteness, we suppose that a1 < a2 (and then b1 > b2 ). We can resolve M(j,k) by free A-modules by the following exact sequence: Ψ

Ψ

0 1 Aα1 ⊕ Aα2 −→ M(j,k) −→ 0 0 −→ Aβ −→

with Ψ0 (β) = y a2 −a1 α1 − z b1 −b2 α2 and Ψ1 (α1 ) = m1 and Ψ1 (α2 ) = m2 . We make this exact sequence graded by declaring that α1 has the same degree as m1 , that α2 has the same degree as m2 and that β has the same degree as y a2 z b1 . Then by the additivity of Hilbert series H(·, λ) we find that H(M(j,k) , λ) = −H(Aβ, λ) + H(Aα1 ⊕ Aα2 , λ) = −H(A, λ)λa2 +b1 + H(A, λ)λa1 +b1 + H(A, λ)λa2 +b2 = H(A, λ)(−λa2 +b1 + λa1 +b1 + λa2 +b2 ) As an example, consider M(1,2) which is generated by m1 = yz 12 and m2 = y 5 z 2 . We can resolve M(1,2) by free A-modules as follows: Ψ

Ψ

0 1 Aα1 ⊕ Aα2 −→ M(1,2) −→ 0 0 −→ Aβ −→

with Ψ0 (β) = y 4 α1 − z 10 α2 and Ψ1 (α1 ) = yz 12 and Ψ1 (α2 ) = y 5 z 2 . Thus H(M(1,2) , λ) = H(A, λ)(−λ17 + λ13 + λ7 ). This argument yields the following Hilbert series: H(M(1,0) , λ) = H(A, λ)(−λ13 + λ11 + λ3 ), H(M(0,2) , λ) = H(A, λ)(−λ14 + λ12 + λ4 ), H(M(1,2) , λ) = H(A, λ)(−λ17 + λ13 + λ7 ), H(M(0,4) , λ) = H(A, λ)(−λ18 + λ14 + λ8 ), H(M(1,4) , λ) = H(A, λ)(−λ17 + λ15 + λ7 ),

210

13 Using SAGBI Bases to Compute Rings of Invariants

H(M(0,6) , λ) = H(A, λ)(−λ18 + λ16 + λ8 ), H(M(1,6) , λ) = H(A, λ)(−λ19 + λ17 + λ9 ), H(M(0,8) , λ) = H(A, λ)(−λ20 + λ18 + λ10 ). Using Equation (13.0.1) we have the following calculation of H(M, λ). First, let q(λ) = (1+λ3 +λ4 +2λ7 +2λ8 +2λ9 +λ10 +2λ11 +2λ12 +λ13 +λ14 +2λ15 +2λ16 ) . Then H(M, λ) = q(λ)H(A, λ) =

(1 − λ)(1 −

q(λ) . − λ10 )(1 − λ11 )

λ2 )(1

Simplifying this yields 1 + λ3 + λ7 + λ8 + 2λ9 + 2λ10 (1 − λ)(1 − λ2 )(1 − λ4 )(1 − λ11 ) 1 − 2λ + 2λ2 − λ3 + λ4 − 2λ5 + 2λ6 = (1 − λ)3 (1 − λ11 )

H(M, λ) =

= H(F[V4 ]C11 , λ). Thus the monomials generated by the elements of Γ ∪{x, y 2 , z 10 , w11 } form a vector space basis for LT (F[V4 ]C11 ) and so B is a SAGBI basis (and thus a generating set) for F[V4 ]C11 . As stated above, this method works entirely similarly to verify a SAGBI basis for F[V4 ]Cp for general p ≥ 5. The reader may see all the details in [98] from which this example was adapted.

14 Ladders

We consider a group G with a normal subgroup N . If σ ∈ G and τ ∈ N , then τ σ = στ  for some τ  ∈ N by normality. Therefore for f ∈ F[V ]N , we have τ · (σ · f ) = τ σ · f = στ  · f = σ · f and thus σ · f ∈ F[V ]N . This shows that G acts on F[V ]N . Clearly (F[V ]N )G = F[V ]G . Since N acts trivially on F[V ]N , in fact, G/N acts on F[V ]N and (F[V ]N )G/N = F[V ]G . We have seen this in detail in Lemma 1.10.1. This description suggests that we may compute F[V ]G by first computing F[V ]N and then computing the G/N invariants in the ring of F[V ]N . Thus we may work with the two smaller groups N and G/N . In characteristic zero, this method is very powerful. See for example the work of Littelmann [76], Popov [89], Wehlau [110]. A major difficulty with this technique is that we must find the invariants of the G/N action on a ring which is usually not a polynomial ring. If G/N is a linearly reductive group, then this difficulty can be surmounted by replacing F[V ]N by a polynomial ring as follows. The space of decomposable invariants, 2 N D := (F[V ]N + ) is a G/N -stable ideal in F[V ] . If G/N is linearly reductive, we may choose a G/N stable complement, Q to D: F[V ]N = D ⊕ Q. A set {f1 , . .. , fr } of N-invariants is a generating set for F[V ]N if and only if the images f1 , . . . , fr span F[V ]N /D. Thus we may choose generators for F[V ]N by taking a basis {f1 , . . . , fm } of Q. We then consider the dual of Q, Q∗ as a G/N representation space. More precisely, we introduce indeterminates y1 , . . . , ym of degree 1 and define a G/N action on F[y1 , . . . , ym ] as follows. For σ ∈ G/N we put m  σ αi,j yj σ · yi = j=1

where

σ αi,j

∈ F is defined by

H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, 211 Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9 14, © Springer-Verlag Berlin Heidelberg 2011

212

14 Ladders

σ · fi =

m 

σ αi,j fj .

j=1

The point here is that F[y1 , . . . , ym ] is a polynomial ring of Krull dimension m whereas there are usually algebraic relations among the functions f1 , . . . , fm . There is a natural G/N -equivariant algebra surjection ρ : F[W ] ∼ = F[y1 , . . . , ym ] → F[f1 , . . . , fm ] = F[V ]N carrying yi to fi where W ∼ = Q is the m dimensional G/N -representation dual to spanF {y1 , . . . , ym }. Since G/N is linearly reductive, ρ restricts to the algebra surjection ρG/N : F[W ]G/N ∼ = F[y1 , . . . , ym ]G/N → F[f1 , . . . , fm ]G/N = F[V ]G . Thus to compute F[V ]G , it suffices to compute F[V ]N and F[W ]G/N . If G/N is not reductive, the above programme requires two modifications. The first change is not too difficult. The construction of W must be done more carefully. It no longer suffices to lift a minimal generating set f1 , . . . , fm of F[V ]N . We require a set of generators h1 , . . . , hr whose span is a G/N stable vector space. To do this, we begin with a set of generators (which need not  to be the vector space be minimal), {f1 , . . . , fm } for F[V ]N . We define Q  := span {σ · fi | σ ∈ G, i = 1, . . . , m} Q F  Then we use indeterminates and choose a vector space basis h1 , . . . , hr for Q. y1 , . . . , yr as before: for σ ∈ G, we define σ · yi =

r 

σ αi,j yj

j=1

where σ · hi =

r 

σ αi,j hj .

j=1

∼ Q.  Since this construction produces Thus F[W ] = F[y1 , . . . , yr ] where W = a larger G/N -module W than in the reductive case, the computation of F[W ]G/N may be expected to be correspondingly harder. The second difficulty we must overcome, if G/N is not reductive, is more serious. Although ρ : F[y1 , . . . , ym ] → F[f1 , . . . , fm ] = F[V ]N is an algebra surjection, the restriction ρG/N : F[y1 , . . . , ym ]G/N → F[f1 , . . . , fm ]G/N = F[V ]G is usually not surjective. Before discussing how to deal with this, we give an example of an algebra surjection whose restriction is not surjective.

14.1 Group Cohomology

213

Example 14.0.1. We take F a field of characteristic p and G = Cp with generator σ. With F[V2 ] = F[x, y] and F[V1 ] = F[u] we have σ · y = y + x, σ · x = x and σ · u = u. Consider the surjective algebra homomorphism ρ : F[V2 ] → F[V1 ] given by ρ(y) = u and ρ(x) = 0. This surjection is Cp -equivariant since σ · ρ(y) = σ · u = u = ρ(y + x) = ρ(σ · y) σ · ρ(x) = σ · 0 = 0 = ρ(σ · x).

and

However, at the level of invariants, the map is not surjective: ρCp : F[V2 ]Cp = F[y p − xp−1 y, x] → F[V1 ] = F[u] where ρCp (y p − xp−1 y) = up and ρCp (x) = 0. Now we return to the question of overcoming this difficulty. We use group cohomology to measure the extent to which ρG/N fails to be surjective.

14.1 Group Cohomology A good reference for group cohomology may be found in Benson’s book [8]. We will very briefly recall the definitions needed for cyclic groups. Let G = Cm be a cyclic group of order m with generator σ. Define two maps, Tr and Δ by Tr(x) = x + σ(x) + σ 2 (x) + · · · + σ m−1 (x) and Δ(x) = σ(x) − x. With this notation we construct a periodic projective resolution of the trivial Cm module, F as follows. Put Fi ∼ = FG for all i and define ∂i : Fi → Fi−1 by  Δ if i is odd, ∂i = Tr if i is even. Thus we have Tr

Δ

Tr

Δ

Tr

Δ

· · · → FG → FG → FG → FG → FG → FG with F0 /Im(∂1 ) = FG/Im(Δ) ∼ = F. Applying the functor HomFG (·, M ) and taking the cohomology of the resulting complex we get H 0 (Cm ; M ) = Kernel(Δ|M ) = M Cm , Kernel(Tr |M ) , H 1 (Cm ; M ) = Im( Δ|M ) Kernel( Δ|M ) , H 2 (Cm ; M ) = Im(Tr |M )

214

14 Ladders

and, for i > 0, H 2i+1 (Cm ; M ) = H 1 (Cm ; M ) and H 2i (Cm ; M ) = H 2 (Cm ; M ). Since the cohomology functor is additive, we see that a Cm -module decomposition of M gives a vector space decomposition of H i (Cm ; M ) in terms of H i (Cm ; V ) with V indecomposable. We now specialize to the case where m = p, is the characteristic of F. By the above, we see that it is important to understand H i (Cp ; Vn ). The reader may wish to refer back to §7.1. Consider any triangular basis {e1 , e2 , . . . , en } of Vn . Clearly H 0 (Cp ; Vn ) is a one dimensional vector space spanned by e1 . Using p−1  the fact that Δp−1 = (σ − 1)p−1 = =0 σ = Tr, we see that H 1 (Cp ; Vp ) = H 2 (Cp ; Vp ) = 0 while, for n < p, we have H 1 (Cp ; Vn ) = {en }

and

H (Cp ; Vn ) = {e1 } . 2

The first equality follows since Δ(Vn ) = Vn−1 and Tr(Vn ) = {0}. The second equality follows similarly since e1 spans Kernel(Δ) and Im(Tr |Vn ) = 0.

14.2 Cohomology and Invariant Theory Recall the situation at the beginning of this chapter with G/N acting on F[V ]N . Choose a set of generators {f1 , f2 , . . . , fm } for F[V ]N and a polynomial algebra A = F[y1 , y2 , . . . , ym ] equipped with a G/N -action such that the map ρ : A → F[V ]N given by ρ(yi ) = fi is G/N equivariant. We recall the notation W for the representation of G/N given by {y1 , y2 , . . . , ym } so that A = F[W ]. We let J denote the ideal in A which is the kernel of ρ. The short exact sequence of A-modules γ

ρ

0 → J −→ A −→ F[V ]N → 0 induces a long exact sequence in group cohomology γ G/N

ρG/N

δ

γ1

ρ1

0 → J G/N → AG/N → F[V ]G → H 1 (G/N ; J) → H 1 (G/N ; A) → . . . . Of course, all the maps in this long exact sequence are AG/N -module maps. From the sequence, we see that F[V ]G ∼ = Im(δ) ker(δ) and therefore, we have the following isomorphisms of vector spaces. F[V ]G ∼ = Im(ρG/N ) ⊕ ker(γ 1 ). = ker(δ) ⊕ Im(δ) ∼

14.2 Cohomology and Invariant Theory

215

This shows that F[V ]G is generated by module generators of the two F[W ]G/N modules: Im(ρG/N ) and ker(γ 1 ). Therefore, in order to find a generating set for F[V ]G in this situation, we have the additional work of computing F[W ]G/N -module generators of ker(γ 1 ). In practice, in our experience, doing so is a hard problem. Indeed, even computing H 1 (G/N ; J) and H 1 (G/N ; F[W ]) seems impractical unless G/N is cyclic. We finish this section by giving three examples. Example 14.2.1. We let F denote the finite field of order p2 and we choose an element ω ∈ F with ω not in the prime field Fp . Thus w generates F over Fp . We consider the subgroup of GL(3, F) generated by the three matrices. ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ 100 100 100 τ −1 = ⎝ 0 1 0⎠ , μ−1 = ⎝0 1 0⎠ , σ−1 = ⎝1 1 0⎠ , 011 ω01 101 Then G is a group of order p3 . Let V denote the three dimensional F vector space on which G is acting and let {x, y, z} denote the basis of V ∗ dual to the ordinary basis. Let H be the subgroup of order p2 generated by σ and μ. Since the index of H in G is p, H is normal in G. Note that H is a Nakajima group and thus we have 2 2 F[V ]H = F[x, NH (y), NH (z)] = F[x, y, z p − xp −1 z]. H 2 Hence the space of indecomposables Q(F[V ]H ) = (F[V ]H + )/(F[V ]+ ) is a three dimensional vector space with basis the images of

2 2 x, y, z p − xp −1 z

in the quotient. Now we consider the action of G/H ∼ = Cp on Q(F[V ]H ). The image of τ generates G/H, and we write Δ := τ − 1. Then Δ(x) = 0, Δ(y) = x and 2 2 Δ(NH (z)) = y p − xp −1 y. Hence Q is not Cp stable and so we extend Q to  with basis the four dimensional space Q 2

2

{y, x, yp − xp

−1

2

2

y, z p − xp

−1

z}.

Dualizing and lifting we take W = V2 ⊕ V2 . We define A := F[W ] = F[X1 , Y1 , X2 , Y2 ] where Δ(Xi ) = 0 and Δ(Yi ) = Xi for i = 1, 2. Define ρ : A → F[V ]H by ρ(Y1 ) = y, ρ(X1 ) = x, ρ(Y2 ) = NH (z) 2 2 and ρ(X2 ) = Δ(ρ(Y2 )) = y p − xp −1 y. Then ρ is a Cp -equivariant algebra surjection. The short exact sequence of A-modules γ

ρ

0 → J −→ A −→ F[V ]H → 0

216

14 Ladders

induces a long exact sequence in group cohomology γ G/H

ρG/H

γ1

δ

ρ1

0 → J G/H → AG/H → F[V ]G → H 1 (G/H; J) → H 1 (G/H; A) → . . . . We will show that ρG/N is surjective. By exactness, this is equivalent to showing that γ 1 : H 1 (G/H; J)→H 1 (G/H; A) is injective. 2 2 Define N := Y1p − X1p −1 Y1 . Then the kernel of ρ, J, is the principal 2 2 ideal generated by Y1p − X1p −1 Y1 − X2 = N − X2 . We write [f ]A to denote the cohomology class in H 1 (G/H; A) represented by f ∈ A. Similarly, [f ]J denotes the cohomology class in H 1 (G/H; J) represented by f ∈ J. Suppose [f  (X2 −N )]J ∈ ker γ 1 . Since [f  (X2 −N )]A = 0, we have f  (X2 − N ) ∈ Δ(A) and so we may write f  (X2 − N ) = Δ(f ) for some f ∈ A. Dividing f by N considered as a polynomial in Y1 we have f = qN + r where degY1 (r) < p2 . Similarly, f  = q  N + r where degY1 (r ) < p2 . Thus f  (N − X2 ) = Δ(f ) = Δ(q)N + Δ(r) and f  (N − X2 ) = (q  N + r  )(N − X2 ) = N (q  N + r  − X2 ) − r X2 . Thus q  N + r − X2 = Δ(q) ∈ Δ(A). Hence f  = q  N + r = Δ(q + Y2 ) ∈ Δ(A) and thus, f  (N − X2 ) = Δ((q + Y2 )(N − X2 )) ∈ Δ(J). Therefore, [f  (N − X2 )]J = 0 and thus γ 1 is injective. This shows that ρG/N : ACp → F[V ]G is surjective. We define f1 = x, f2 = NG (y) = y p − xp−1 y, f3 = ρ(X2 Y1 − X1 Y2 ) 2

2

= x(z p − xp 2

= yp

+1

−1

2

2

2

− xz p − xp

G

f4 = N (z) = z

3

p

2

z) − (y p − xp −1 2

−1

2

y)y

y + xp z,

and

+ ...

Then F[V ]G is generated by ρ(X1 ) = x = f1 , ρ(N

Cp

(Y1 )) = f2 ,

ρ(N

Cp

(Y2 )) = f4 .

2

2

ρ(X2 ) = y p − xp

−1

ρ(X1 Y2 − X2 Y1 ) = f3 ,

Hence F[V ]G = F[f1 , f2 , f3 , f4 ]. Example 14.2.2. Consider the matrix

2

y = f2p + f1p and

−p

f2 ,

14.2 Cohomology and Invariant Theory

⎛

217

⎞

100 σ −1 = ⎝1 1 0⎠ 011 acting on the three dimensional vector space V3 over F of characteristic 2. It is easy to see that σ generates a cyclic group, G, of order 4, and we write F[V3 ] = F[x, y, z]. Then σ · x = x, σ · y = y + x, σ · z = z + y.

and

We consider the normal subgroup N of G generated by σ 2 . We see immediately that as FN -modules, V3 ∼ = V2 ⊕ V1 , where V2 has basis {x, z} and V1 has basis {y}. Therefore, as FN -modules, we have F[V3 ] = F[V2 ] ⊗ F[V1 ] = F[x, z] ⊗ F[y]. Therefore,

F[V3 ]N ∼ = F[V2 ⊕ V1 ]N = F[x, z 2 − xz, y].

N 2 Thus the space of indecomposables Q(F[V3 ]N ) = F[V3 ]N + /(F[V3 ]+ ) is the 2 three dimensional vector space with basis {x, y, z + xz}. Next, we consider the action of G/N ∼ = C2 on Q(F[V3 ]N ). Let τ denote a generator of G/N . Then

τ · x = x, τ · y = y + x,

and

τ · (z + xz) = z + xz + y 2 + xy. 2

2

Hence the abstract vector space spanF {x, y, z 2 +xz, z 2 +xz} is not G/N stable  with basis and we must instead consider the abstract 4 dimensional space, Q, {x, y, z 2 + xz, y2 + xy} which is G/N -stable since τ · (y2 + xy) = y 2 + xy. We dualize and lift to obtain W ∼ = V2 ⊕ V2 and we write F[W ] = F[a1 , b1 , a2 , b2 ] where τ · ai = ai , and τ · b i = bi + a i Define ρ : F[W ] → F[V3 ]N

218

14 Ladders

by ρ(a1 ) = x, ρ(b1 ) = y, ρ(a2 ) = y2 + xy,

and

2

ρ(b2 ) = z + xz. Then ρ is a G/N -equivariant surjection. As usual, the short exact sequence of A := F[W ]-modules γ

ρ

0 → J −→ A −→ F[V3 ]N → 0 induces a long exact sequence in group cohomology γ G/N

ρG/N

δ

γ1

ρ1

0 → J G/N → AG/N → F[V3 ]G → H 1 (G/N ; J) → H 1 (G/N ; A) → . . . . We claim that in this case, the map ρG/N is surjective. By exactness, we may prove this by showing that γ 1 : H 1 (G/N ; J)→H 1 (G/N ; F[W ]) is injective. Since p = 2, we have that TrG/N = τ + Id = τ − Id = Δ. Therefore Kernel(Tr |M ) , Im( Δ|M ) Kernel(Δ|M ) = Im( Tr |M )

H 1 (G/N ; M ) =

As usual, for f ∈ F[W ]Cp we let [f ]F[W ] denote the cohomology class in H 1 (G/N ; F[W ]) represented by f . Similarly if f ∈ J Cp , we denote by [f ]J the cohomology class f represents in H 1 (G/N ; J). To see that γ 1 is injective, consider an element f ∈ J Cp such that γ 1 ([f ]F[W ] ) = 0. Thus f ∈ TrG N (F[W ]).   (f ) for some f ∈ F[W ]. Write f = TrG N Note that F[W ] has Krull dimension 4 and that F[V ]N has Krull dimension 3. We note that ρ(a2 ) = ρ(b21 + a1 b1 ), and so J is the principal ideal generated by the element r := a2 + b21 + a1 b1 . Since f ∈ J, we may write f = f  r for some f  ∈ F[W ]. Since f ∈ Im Tr, it must vanish on G/N W = {(β2 , 0, β1 , 0) | β2 , β1 ∈ F}. Thus 0 = f  (β2 , 0, β1 , 0) = f  (β2 , 0, β1 , 0) · r(β2 , 0, β1 , 0) = f  (β2 , 0, β1 , 0)β12 .

14.2 Cohomology and Invariant Theory

219

Therefore f  must vanish on the set {(β2 , 0, β1 , 0) | β1 = 0, β2 , β1 ∈ F}. Hence, by continuity, f  must vanish on

Therefore f  ∈



{(β2 , 0, β1 , 0) | β2 , β1 ∈ F} = W

G/N

.

Im TrG N . Since W is equivalent to a permutation represen-

tation, Proposition 9.0.13 implies that f  ∈ Im TrG N . Thus there exists an ele   ment f  ∈ F[W ] with TrG (f ) = f . This shows that f = f  r = TrG N N (f )r = G G  TrN (f r) ∈ TrN (J). Therefore [f ]J = 0. Thus we have shown that γ 1 is injective and so ρG/N is surjective. Now F[W ]G/N = F[a1 , b1 , a2 b1 + a1 b2 , b21 + a1 b1 , b22 + a2 b2 ] and F[V3 ]G = ρ(F[W ]G/N ) = F[f1 , f2 , f3 , f4 ] where f1 = x, f2 = y 2 + xy, f3 = y 3 + xy 2 + xz 2 + x2 z, 4

2 2

2 2

and 2

f4 = z + x z + y z + xyz + xy2 z + x2 yz. In particular, we see that F[V3 ]G is a hypersurface. Furthermore, it is straightforward to verify that the relation among the generators is given by f23 + f32 + f1 f2 f3 + f12 f4 = 0 . Example 14.2.3. Let F4 = {0, 1, ω, ω 2 } be the field of order 4 where ω satisfies ω 2 + ω + 1 = 0. Consider the subgroup G of GL(3, F4 ) generated by the 2 matrices ⎛ ⎛ ⎞ ⎞ 100 100 σ −1 = ⎝1 1 0⎠ , τ −1 = ⎝ 0 1 0⎠ 011 ω01 Let V denote the three dimensional F4 vector space on which G acts with {x, y, z} the basis of V ∗ dual to the standard basis. Notice that the subgroup N generated by σ is the group we considered in Example 14.2.2. Thus N has order 4 and F4 [V ]N = F4 [f1 , f2 , f3 , f4 ] where f1 = x, f2 = y 2 + xy, f3 = y 3 + xy 2 + xz 2 + x2 z, f4 = z 4 + x2 z 2 + y 2 z 2 + xyz 2 + xy 2 z + x2 yz.

220

14 Ladders

Furthermore, τ 2 = I3 and thus G has order 8 and N is normal in G. We will use a ladder to study the ring of G invariants: F4 [V ]G = (F4 [V ]N )G/N where G/N ∼ = C2 . Write Δ = τ − 1. The action of G/N on F4 [V ]N is given by Δ(f1 ) = 0, Δ(f2 ) = 0, Δ(f3 ) = f13 , Δ(f4 ) =

f12 (f2

and + f12 ) .

Thus we define W = 2 V1 ⊕ 2 V2 where W ∗ has basis {a1 , a2 , a3 , b3 , a4 , b4 } with τ (ai ) = ai for i = 1, 2, 3, 4 and τ (bi ) = bi + ai fori = 3, 4 We define ρ : A := F4 [W ] → F4 [V ]N by ρ(a1 ) = f1 , ρ(a2 ) = f2 , ρ(b3 ) = f3 , ρ(a3 ) = Δ(f3 ) = f13 , ρ(b4 ) = f4 and ρ(a4 ) = Δ(f4 ) = f14 + f12 f2 . Thus ρ is a G/N -equivariant surjection. Again the short exact sequence of A-modules γ

ρ

0 → J −→ A −→ F4 [V ]N → 0 induces a long exact sequence in group cohomology γ G/N

ρG/N

δ

γ1

ρ1

0 → J G/N → AG/N → F4 [V ]G → H 1 (G/N ; J) → H 1 (G/N ; A) → . . . . In this case, we will see that ρG/N is not surjective. We define f5 := y 5 + xz 4 + x3 y 2 + x4 z. Since Δ(f5 ) = x(ωx)4 + x4 (ωx) = 0, we see that f5 is G-invariant. It is easy to see that f5 does not lie in the image of ρG/N as follows. We have ρG/N (F4 [W ]G/N ) = ρG/N (F4 [a1 , a2 , b23 + a3 b3 , b24 + a4 b4 , a3 b4 + a4 b3 ]) = F4 [f1 , f2 , f32 + f13 f3 , f42 + f14 f4 + f12 f2 f4 , f13 f4 + f14 f3 + f12 f2 f3 ].

14.2 Cohomology and Invariant Theory

221

Note that fi has degree i for all i = 1, 2, . . . , 5 and thus degree considerations show that if f5 ∈ ρG/N (F4 [W ]G/N ), then f5 ∈ F4 [f1 , f2 ]5 . Since every element of F4 [f1 , f2 ]5 is divisible by f1 = x and the lead monomial of f5 is y 5 we see that f5 ∈ / F4 [f1 , f2 ]5 . The ideal J is generated by the 3 relations a32 + b23 + a1 a2 b3 + a21 b4 a3 + a31 a4 +

a21 a2

and + a41 .

Consider f = (a2 + a21 )(a3 + a31 ) + a1 (a4 + a21 a2 + a41 ) = a2 a3 + a21 a3 + a1 a4 ∈ J. Note that f = Δ(f  ) ∈ Δ(F4 [W ]) where f  = a2 b3 + a21 b3 + a1 b4 . However, f ∈ / Δ(J). Therefore f represents a non-zero cohomology class [f ]J ∈ H 1 (G/N ; J) with γ 1 ([h]J ) = 0 ∈ H 1 (G/N ; F4 [W ]). This corresponds to the invariant ρ(g) = f2 f3 + f12 f3 + f1 f4 = f5 ∈ F4 [W ]G . With a little more work the reader may show that the F4 [W ]G/N -module, ker γ 1 is generated by the element just considered [a2 a3 + a21 a3 + a1 a4 ]J . Thus F4 [V ]G is generated by ρ(F4 [W ]H ) together with f5 . From this it easily follows that F4 [V ]G = F4 [f1 , f2 , f5 , f8 ] where f8 = ρ(NG/H (b4 )) = z 8 + y 4 z 4 + xy 7 + x2 y 2 z 4 + x3 y 4 z + x3 y 5 x5 y 2 z + x6 z 2 . Remark 14.2.4. Note that the representation considered in Example 14.2.3 is the same representation as that considered in Example 14.2.1 when p = 2. However, the choice of ladder (i.e., the tower of normal subgroups used) is different in the two examples.

References

1. William W. Adams and Philippe Loustaunau, An introduction to Gr¨ obner bases, Graduate Studies in Mathematics, vol. 3, American Mathematical Society, Providence, RI, 1994. MR 1287608 (95g:13025) 2. Gert Almkvist, Representations of Z/pZ in characteristic p and reciprocity theorems, J. Algebra 68 (1981), no. 1, 1–27. MR 604290 (82k:14047) 3. Gert Almkvist and Robert Fossum, Decomposition of exterior and symmetric powers of indecomposable Z/pZ-modules in characteristic p and relations to invariants, S´eminaire d’Alg`ebre Paul Dubreil, 30`eme ann´ee (Paris, 1976– 1977), Lecture Notes in Math., vol. 641, Springer, Berlin, 1978, pp. 1–111. MR 499459 (81b:14024) 4. Emil Artin, Galois theory, second ed., Dover Publications Inc., Mineola, NY, 1998, Edited and with a supplemental chapter by Arthur N. Milgram. MR 1616156 (98k:12001) 5. M. F. Atiyah and I. G. Macdonald, Introduction to commutative algebra, Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont., 1969. MR 39:4129 6. D. J. Benson, Polynomial invariants of finite groups, London Mathematical Society Lecture Note Series, vol. 190, Cambridge University Press, Cambridge, 1993. MR 94j:13003 , Representations and cohomology. I, second ed., Cambridge Studies 7. in Advanced Mathematics, vol. 30, Cambridge University Press, Cambridge, 1998, Cohomology of groups and modules. MR 1634407 (99f:20001b) 8. , Representations and cohomology. II, second ed., Cambridge Studies in Advanced Mathematics, vol. 31, Cambridge University Press, Cambridge, 1998, Cohomology of groups and modules. MR 1634407 (99f:20001b) 9. Marie-Jos´e Bertin, Anneaux d’invariants d’anneaux de polynomes, en caract´eristique p, C. R. Acad. Sci. Paris S´er. A-B 264 (1967), A653–A656. MR 0215826 (35 #6661) 10. W. Bosma, J. Cannon, and C. Playous, The magma algebra system. i. the user language, J. Symbolic Comput. 24 (1997), no. 3–4, 235–265. ´ ements de math´ematique. I: Les structures fondamentales 11. N. Bourbaki, El´ de l’analyse. Fascicule XI. Livre II: Alg`ebre. Chapitre 4: Polynomes et fractions rationnelles. Chapitre 5: Corps commutatifs, Deuxi`eme ´edition.

H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9, © Springer-Verlag Berlin Heidelberg 2011

223

224

12. 13. 14.

15.

16.

17.

18. 19. 20. 21.

22. 23. 24. 25.

26.

27. 28. 29.

30.

References Actualit´es Scientifiques et Industrielles, No. 1102, Hermann, Paris, 1959. MR 0174550 (30 #4751) Abraham Broer, Remarks on invariant theory of finite groups, Unpublished manuscript, 1997. , Differents in modular invariant theory, Transform. Groups 11 (2006), no. 4, 551–574. MR 2278139 (2007h:13007) Winfried Bruns and J¨ urgen Herzog, Cohen-Macaulay rings, Cambridge Studies in Advanced Mathematics, vol. 39, Cambridge University Press, Cambridge, 1993. MR 1251956 (95h:13020) Roger M. Bryant and Gregor Kemper, Global degree bounds and the transfer principle for invariants, J. Algebra 284 (2005), no. 1, 80–90. MR 2115005 (2005i:13007) H. E. A. Campbell and Jianjun Chuai, On the invariant fields and localized invariant rings of p-groups, Quarterly Journal of Mathematics 10 (2007), no. 10.1093/qmath/ham011, 1–7. H. E. A. Campbell, A. V. Geramita, I. P. Hughes, R. J. Shank, and D. L. Wehlau, Non-Cohen-Macaulay vector invariants and a Noether bound for a Gorenstein ring of invariants, Canad. Math. Bull. 42 (1999), no. 2, 155–161. MR 2000b:13005 H. E. A. Campbell, I. Hughes, and R. D. Pollack, Rings of invariants and pSylow subgroups, Canad. Math. Bull. 34 (1991), no. 1, 42–47. MR 92h:13008 H. E. A. Campbell and I. P. Hughes, Vector invariants of U2 (Fp ): a proof of a conjecture of Richman, Adv. Math. 126 (1997), no. 1, 1–20. MR 98c:13007 H. E. A. Campbell and P. S. Selick, Polynomial algebras over the Steenrod algebra, Comment. Math. Helv. 65 (1990), no. 2, 171–180. MR 1057238 (91f:55006) H. E. A. Campbell, R. J. Shank, and David L. Wehlau, Vector invariants for the two dimensional modular representation of a cyclic group of prime order, Advances in Math. 225 (2010), no. 2, 1069–1094. Claude Chevalley, Invariants of finite groups generated by reflections, Amer. J. Math. 77 (1955), 778–782. MR 17,345d Jianjun Chuai, Two-dimensional vector invariant rings of abelian p-groups, J. Algebra 266 (2003), no. 1, 362–373. MR 2004e:13010 H. S. M. Coxeter, The product of the generators of a finite group generated by reflections, Duke Math. J. 18 (1951), 765–782. MR 0045109 (13,528d) Charles W. Curtis and Irving Reiner, Representation theory of finite groups and associative algebras, Wiley Classics Library, John Wiley & Sons Inc., New York, 1988, Reprint of the 1962 original, A Wiley-Interscience Publication. MR 1013113 (90g:16001) Harm Derksen and Gregor Kemper, Computational invariant theory, Invariant Theory and Algebraic Transformation Groups, I, Springer-Verlag, Berlin, 2002, Encyclopaedia of Mathematical Sciences, 130. MR 2003g:13004 , Computing invariants of algebraic groups in arbitrary characteristic, Adv. Math. 217 (2008), no. 5, 2089–2129. MR 2388087 (2009a:13005) L. E. J. Dickson, Theorems on the residues of multinomial coefficients with respect to a prime modulus, Q. J. Pure Appl. Math. 33 (1902), 378–384. Jan Draisma, Gregor Kemper, and David Wehlau, Polarization of separating invariants, Canad. J. Math. 60 (2008), no. 3, 556–571. MR 2414957 (2009c:13011) Emilie Dufresne, Separating invariants, Ph.D. thesis, Queen’s University, Kingston, Ontario, Canada, 2008, defended.

References 31. 32. 33. 34.

35. 36. 37. 38.

39. 40.

41.

42.

43.

44. 45.

46.

47.

48.

49.

225

, Separating invariants of finite reflections groups, Advances in Math. 221 (2009), no. 6, 1979–1989. David S. Dummit and Richard M. Foote, Abstract algebra, third ed., John Wiley & Sons Inc., Hoboken, NJ, 2004. MR 2286236 (2007h:00003) David Eisenbud, Commutative algebra, Graduate Texts in Mathematics, vol. 150, Springer-Verlag, New York, 1995. MR 97a:13001 , Commutative algebra, Graduate Texts in Mathematics, vol. 150, Springer-Verlag, New York, 1995, With a view toward algebraic geometry. MR 1322960 (97a:13001) Geir Ellingsrud and Tor Skjelbred, Profondeur d’anneaux d’invariants en caract´eristique p, Compositio Math. 41 (1980), no. 2, 233–244. MR 82c:13015 Jonathan Elmer and Peter Fleischmann, On the depth of modular invariant rings for the groups Cp × Cp , Progress in Mathematics 278 (2009), 45–62. Charles S. Fisher, The death of a mathematical theory: a study in the sociology of knowledge, Arch. History Exact Sci. 3 (1966), no. 2, 137–159. MR 0202546 Peter Fleischmann, Relative trace ideals and Cohen-Macaulay quotients of modular invariant rings, Computational methods for representations of groups and algebras (Essen, 1997), Progr. Math., vol. 173, Birkh¨ auser, Basel, 1999, pp. 211–233. MR 1714612 (2000j:13007) , The Noether bound in invariant theory of finite groups, Adv. Math. 156 (2000), no. 1, 23–32. MR 1800251 (2001k:13009) Peter Fleischmann, Gregor Kemper, and Chris Woodcock, Homomorphisms, localizations and a new algorithm to construct invariant rings of finite groups, J. Algebra 309 (2007), no. 2, 497–517. MR 2303190 Peter Fleischmann and R. James Shank, The relative trace ideal and the depth of modular rings of invariants, Arch. Math. (Basel) 80 (2003), no. 4, 347–353. MR 2004e:13012 John Fogarty, On Noether’s bound for polynomial invariants of a finite group, Electron. Res. Announc. Amer. Math. Soc. 7 (2001), 5–7 (electronic). MR 1826990 (2002a:13002) Robert M. Fossum and Phillip A. Griffith, Complete local factorial rings which ´ are not Cohen-Macaulay in characteristic p, Ann. Sci. Ecole Norm. Sup. (4) 8 (1975), no. 2, 189–199. MR 0382257 (52 #3142) Manfred G¨ obel, Computing bases for rings of permutation-invariant polynomials, J. Symbolic Comput. 19 (1995), no. 4, 285–291. MR 96f:13006 , A constructive description of SAGBI bases for polynomial invariants of permutation groups, J. Symbolic Comput. 26 (1998), no. 3, 261–272. MR 1633927 (99f:13002) , The optimal lower bound for generators of invariant rings without finite SAGBI bases with respect to any admissible order, Discrete Math. Theor. Comput. Sci. 3 (1999), no. 2, 65–70 (electronic). MR 1695195 (2001b:13008) , Rings of polynomial invariants of the alternating group have no finite SAGBI bases with respect to any admissible order, Inform. Process. Lett. 74 (2000), no. 1–2, 15–18. MR 1761193 (2001d:13003) P. Gordan, Beweis dass jede covariante und invariante einer bindren form eine ganze function mit numerischen coefficienten solcher formen ist, J. fur reine u. angew. Math. 69 (1868), 323–354. Paul Gordan, Vorlesungen u ¨ber Invariantentheorie, second ed., Chelsea Publishing Co., New York, 1987, Erster Band: Determinanten. [Vol. I: Determi-

226

50. 51. 52. 53.

54.

55.

56. 57.

58.

59. 60.

61.

62.

63. 64. 65. 66. 67. 68.

References nants], Zweiter Band: Bin¨ are Formen. [Vol. II: Binary forms], Edited by Georg Kerschensteiner. MR 917266 (89g:01034) F. D. Grosshans, Vector invariants in arbitrary characteristic, Transform. Groups 12 (2007), no. 3, 499–514. MR 2356320 Robin Hartshorne, Complete intersections and connectedness, Amer. J. Math. 84 (1962), 497–508. MR 0142547 (26 #116) David Hilbert, Ueber die Theorie der algebraischen Formen, Math. Ann. 36 (1890), no. 4, 473–534. MR 1510634 M. Hochster and John A. Eagon, Cohen-Macaulay rings, invariant theory, and the generic perfection of determinantal loci, Amer. J. Math. 93 (1971), 1020– 1058. MR 46:1787 Ian Hughes and Gregor Kemper, Symmetric powers of modular representations, Hilbert series and degree bounds, Comm. Algebra 28 (2000), no. 4, 2059–2088. MR 2001b:13009 , Symmetric powers of modular representations for groups with a Sylow subgroup of prime order, J. Algebra 241 (2001), no. 2, 759–788. MR 1843324 (2002e:13012) James E. Humphreys, Linear algebraic groups, Springer-Verlag, New York, 1975, Graduate Texts in Mathematics, No. 21. MR 0396773 (53 #633) Victor Kac and Keiichi Watanabe, Finite linear groups whose ring of invariants is a complete intersection, Bull. Amer. Math. Soc. (N.S.) 6 (1982), no. 2, 221– 223. MR 640951 (83h:14042) William M. Kantor, Subgroups of classical groups generated by long root elements, Trans. Amer. Math. Soc. 248 (1979), no. 2, 347–379. MR 522265 (80g:20057) Irving Kaplansky, Commutative rings, revised ed., The University of Chicago Press, Chicago, Ill.-London, 1974. MR 0345945 (49 #10674) Deepak Kapur and Klaus Madlener, A completion procedure for computing a canonical basis for a k-subalgebra, Computers and mathematics (Cambridge, MA, 1989), Springer, New York, 1989, pp. 1–11. MR 1005954 (90g:13001) D. B. Karagueuzian and P. Symonds, The module structure of a group action on a polynomial ring: examples, generalizations, and applications, Invariant theory in all characteristics, CRM Proc. Lecture Notes, vol. 35, Amer. Math. Soc., Providence, RI, 2004, pp. 139–158. MR 2066462 (2005g:13011) Dikran B. Karagueuzian and Peter Symonds, The module structure of a group action on a polynomial ring: a finiteness theorem, J. Amer. Math. Soc. 20 (2007), no. 4, 931–967 (electronic). MR 2328711 G. Kemper and G. Malle, The finite irreducible linear groups with polynomial ring of invariants, Transform. Groups 2 (1997), no. 1, 57–89. MR 98a:13012 Gregor Kemper, A constructive approach to Noether’s problem, Manuscripta Math. 90 (1996), no. 3, 343–363. MR 1397662 (97d:13005) , On the Cohen-Macaulay property of modular invariant rings, J. Algebra 215 (1999), no. 1, 330–351. MR 2000d:13008 , Computing invariants of reductive groups in positive characteristic, Transform. Groups 8 (2003), no. 2, 159–176. MR 2004b:13006 , A course in commutative algebra, first ed., Graduate Texts in Mathematics, vol. 256, Springer, to appear, 2010. Gregor Kemper and Gunter Malle, Invariant fields of finite irreducible reflection groups, Math. Ann. 315 (1999), no. 4, 569–586. MR MR1731462 (2001c:13006)

References 69. 70.

71.

72. 73.

74. 75. 76. 77.

78.

79.

80. 81. 82. 83.

84. 85.

86.

87.

227

, Invariant fields of finite irreducible reflection groups, Math. Ann. 315 (1999), no. 4, 569–586. MR 2001c:13006 Hanspeter Kraft and Claudio Procesi, A primer of invariant theory, notes by G. Boffi, http://www.math.unibas.ch/˜kraft/Papers/KP-Primer.pdf 1 (1982, rev. 2000), 125. Ernst Kunz, Introduction to commutative algebra and algebraic geometry, Birkh¨ auser Boston Inc., Boston, MA, 1985, Translated from the German by Michael Ackerman, With a preface by David Mumford. MR 789602 (86e:14001) Shigeru Kuroda, The infiniteness of the SAGBI bases for certain invariant rings, Osaka J. Math. 39 (2002), no. 3, 665–680. MR 1932287 (2003k:13033) Peter S. Landweber and Robert E. Stong, The depth of rings of invariants over finite fields, Number theory (New York, 1984–1985), Lecture Notes in Mathematics, vol. 1240, Springer, Berlin, 1987, pp. 259–274. MR 88k:13004 Serge Lang, Algebra, second ed., Addison-Wesley Publishing Company Advanced Book Program, Reading, MA, 1984. MR 783636 (86j:00003) , Algebra, third ed., Addison-Wesley Publishing Company Advanced Book Program, Reading, MA, 1993. Peter Littelmann, Koregul¨ are und a ¨quidimensionale Darstellungen, J. Algebra 123 (1989), no. 1, 193–222. MR 1000484 (90e:20039) Martin Lorenz, Multiplicative invariant theory, Encyclopaedia of Mathematical Sciences, vol. 135, Springer-Verlag, Berlin, 2005, Invariant Theory and Algebraic Transformation Groups, VI. MR 2131760 (2005m:13012) I. G. Macdonald, Symmetric functions and Hall polynomials, second ed., Oxford Mathematical Monographs, The Clarendon Press Oxford University Press, New York, 1995, With contributions by A. Zelevinsky, Oxford Science Publications. MR 1354144 (96h:05207) Hideyuki Matsumura, Commutative algebra, second ed., Mathematics Lecture Note Series, vol. 56, Benjamin/Cummings Publishing Co., Inc., Reading, Mass., 1980. MR 575344 (82i:13003) , Commutative ring theory, Cambridge Studies in Advanced Mathematics, vol. 8, Cambridge University Press, Cambridge, 1986. MR 88h:13001 Takehiko Miyata, Invariants of certain groups. I, Nagoya Math. J. 41 (1971), 69–73. MR 0272923 (42 #7804) H. Mui, Modular invariant theory and cohomology algebras of symmetric groups, J. of Fac. Sci. Univ. Tokyo Sect. 1A Math. 22 (1975), 319–369. Haruhisa Nakajima, Modular representations of p-groups with regular rings of invariants, Proc. Japan Acad. Ser. A Math. Sci. 56 (1980), no. 10, 469–473. MR 82a:20016 , Rings of invariants of finite groups which are hypersurfaces, J. Algebra 80 (1983), no. 2, 279–294. MR 85e:20036 Mara D. Neusel, Invariant theory, Student Mathematical Library, vol. 36, American Mathematical Society, Providence, RI, 2007. MR 2280491 (2007m:13007) Mara D. Neusel and Larry Smith, Invariant theory of finite groups, Mathematical Surveys and Monographs, vol. 94, American Mathematical Society, Providence, RI, 2002. MR 1869812 (2002k:13012) P. E. Newstead, Introduction to moduli problems and orbit spaces, Tata Institute of Fundamental Research Lectures on Mathematics and

228

88. 89.

90. 91. 92. 93.

94. 95.

96.

97.

98. 99. 100. 101. 102. 103. 104.

105. 106. 107.

References Physics, vol. 51, Tata Institute of Fundamental Research, Bombay, 1978. MR 546290 (81k:14002) E. Noether, Der endlichkeitssatz der invarianten endlicher linearer gruppen der characteristik p, Nachr. v. d. Ges. d. Wiss. zu G¨ ottingen (1926), 28–35. V. L. Popov, Constructive invariant theory, Young tableaux and Schur functors in algebra and geometry (Toru´ n, 1980), Ast´erisque, vol. 87, Soc. Math. France, Paris, 1981, pp. 303–334. MR 646826 (83i:14040) Claudio Procesi, Lie groups, Universitext, Springer, New York, 2007, An approach through invariants and representations. MR 2265844 (2007j:22016) Zinovy Reichstein, SAGBI bases in rings of multiplicative invariants, Comment. Math. Helv. 78 (2003), no. 1, 185–202. MR 1966757 (2004c:13005) David R. Richman, On vector invariants over finite fields, Adv. Math. 81 (1990), no. 1, 30–65. MR 91g:15020 Lorenzo Robbiano and Moss Sweedler, Subalgebra bases, Commutative algebra (Salvador, 1988), Lecture Notes in Math., vol. 1430, Springer, Berlin, 1990, pp. 61–87. MR 1068324 (91f:13027) Michael Roberts, On the covariants of a binary quantic of the nth degree, Quarterly Journal of Pure and Applied Mathematics 4 (1861), 168–178. Jean-Pierre Serre, Groupes finis d’automorphismes d’anneaux locaux r´ eguliers, Colloque d’Alg`ebre (Paris, 1967), Exp. 8, Secr´etariat math´ematique, Paris, 1968, p. 11. MR 38:3267 M¨ ufit Sezer and R. James Shank, On the coinvariants of modular representations of cyclic groups of prime order, J. Pure Appl. Algebra 205 (2006), no. 1, 210–225. MR 2193198 (2006k:13015) R. J. Shank, Classical covariants and modular invariants, Invariant theory in all characteristics, CRM Proc. Lecture Notes, vol. 35, Amer. Math. Soc., Providence, RI, 2004, pp. 241–249. MR 2005d:13012 R. James Shank, S.A.G.B.I. bases for rings of formal modular seminvariants, Comment. Math. Helv. 73 (1998), no. 4, 548–565. MR 2000a:13016 R. James Shank and David L. Wehlau, Computing modular invariants of pgroups, J. Symbolic Comput. 34 (2002), no. 5, 307–327. MR 2003j:13006 , Noether numbers for subrepresentations of cyclic groups of prime order, Bull. London Math. Soc. 34 (2002), no. 4, 438–450. MR 2003a:13005 G. C. Shephard and J. A. Todd, Finite unitary reflection groups, Canadian J. Math. 6 (1954), 274–304. MR 15,600b Larry Smith, Polynomial invariants of finite groups, Research Notes in Mathematics, vol. 6, A K Peters Ltd., Wellesley, MA, 1995. MR 1328644 (96f:13008) , Polynomial invariants of finite groups. A survey of recent developments, Bull. Amer. Math. Soc. (N.S.) 34 (1997), no. 3, 211–250. MR 98i:13009 Richard P. Stanley, Invariants of finite groups and their applications to combinatorics, Bull. Amer. Math. Soc. (N.S.) 1 (1979), no. 3, 475–511. MR 81a:20015 Bernd Sturmfels, Algorithms in invariant theory, Texts and Monographs in Symbolic Computation, Springer-Verlag, Vienna, 1993. MR 94m:13004 Peter Symonds, On the castelnuovo-mumford regularity of rings of polynomial invariants, preprint (2009), 1–11. N. M. Thi´ery and S. Thomass´e, Convex cones and SAGBI bases of permutation invariants, Invariant theory in all characteristics, CRM Proc. Lecture Notes, vol. 35, Amer. Math. Soc., Providence, RI, 2004, pp. 259–263. MR 2066473 (2005e:13006)

References

229

108. A. Wagner, Collineation groups generated by homologies of order greater than 2, Geom. Dedicata 7 (1978), no. 4, 387–398. MR 512113 (81e:20055) 109. Ascher Wagner, Determination of the finite primitive reflection groups over an arbitrary field of characteristic not 2. I, Geom. Dedicata 9 (1980), no. 2, 239–253. MR 578199 (81g:20096) 110. David Wehlau, Equidimensional representations of 2-simple groups, J. Algebra 154 (1993), no. 2, 437–489. MR 1206131 (93k:14064) 111. David L. Wehlau, The Noether number in invariant theory, C. R. Math. Acad. Sci. Soc. R. Can. 28 (2006), no. 2, 39–62. MR 2257602 (2007h:13008) 112. Hermann Weyl, The classical groups, Princeton Landmarks in Mathematics, Princeton University Press, Princeton, NJ, 1997. MR 98k:01049 113. Clarence Wilkerson, A primer on the Dickson invariants, Proceedings of the Northwestern Homotopy Theory Conference (Evanston, Ill., 1982) (Providence, RI), Contemp. Math., vol. 19, Amer. Math. Soc., 1983, pp. 421–434. MR 85c:55017 114. Yinglin Wu, Rings of invariants of certain modular groups, Ph.D. thesis, Queen’s University, Kingston, Ontario, Canada, 2009, defended. 115. A. E. Zalesski˘ı and V. N. Sereˇzkin, Finite linear groups generated by reflections, Izv. Akad. Nauk SSSR Ser. Mat. 44 (1980), no. 6, 1279–1307, 38. MR 603578 (82i:20060) 116. Oscar Zariski and Pierre Samuel, Commutative algebra. Vol. 1, Springer-Verlag, New York, 1975. MR 52:5641 , Commutative algebra. Vol. II, Springer-Verlag, New York, 1975. 117. MR 52:10706

Index

Almkvist, 76, 78 alternating function, 64 alternating group, 91, 93, 155 Artin, 101 Atiyah, 25, 27 Benson, 1, 8, 106, 107, 213 Bertin, 53, 74 bi-reflection, 9, 170, 172 Bourbaki, 91 Brauer lift, 51 Broer, 8, 173 Bruns, 27 Bryant, 24 C3 , 75 C4 , 53, 72 Campbell, 8, 23, 49, 62, 115, 130, 148, 170 character, 54 Chevalley, 9 Chuai, 62, 130, 173 closed set, 29 Cohen-Macaulay, 6–9, 11, 32–34, 42, 43, 53, 56, 60–62, 69, 71–74, 100, 144, 170–173 cohomology, 170, 213, 214, 216, 218, 220, 221 coinvariants, 99, 135 coordinate ring, 3, 11, 16, 26–28, 88 Coxeter, 9 Cp , 11, 16–18, 20, 60, 75, 76, 88, 105–108, 110, 111, 113, 125, 126,

132–134, 136, 155, 180, 187, 189, 194, 205, 215 Cp r, 106 Dade, 41, 42 depth, 32, 56 Derksen, 1, 191 Dickson, 165 Dickson invariants, 40, 45, 75, 77 Draisma, 191, 201 Dufresne, 191 Dummit, 25, 27, 50 Eagon, 9, 32 Eisenbud, 25, 27 elementary symmetric functions, 7, 13, 15, 40, 44, 45, 53, 61, 69, 72, 90, 91, 102 Elemer, 48 Ellingsrud, 170 equivariant, 214 Fleischmann, 8, 48, 63, 132, 153 Fogarty, 8, 46 Foote, 25, 27, 50 Fossum, 76, 78 free resolution, 36 G¨ obel’s Theorem, 66, 68–70 Galois theory, 75, 130 Geramita, 8, 170 Gordan, 125 Gorenstein, 9, 34 Gr¨ obner basis, 85, 87, 91, 114

H.E.A. Eddy Campbell, D. Wehlau, Modular Invariant Theory, Encyclopaedia of Mathematical Sciences 139, DOI 10.1007/978-3-642-17404-9, © Springer-Verlag Berlin Heidelberg 2011

231

232

Index

graded Nakayama lemma, 35, 36, 42 height, 28, 31, 32, 55, 164, 171, 175, 196 Herzog, 27 Hilbert, 25, 30, 41, 125 Hilbert ideal, 47, 48, 99, 101, 135, 173, 195 Hilbert series, 34, 37, 42, 50–53, 61, 62, 66, 69, 71–78, 85, 99, 101, 113, 135, 138, 139, 205, 207–209 Hilbert syzygy theorem, 36 Hilbert’s Nullstellensatz, 26 Hironaka decomposition, 42, 45, 65, 101, 155, 156, 182 Hochster, 9, 32 homogeneous system of parameters, 30, 41, 42, 44 homology, 141, 150 Hughes, 8, 23, 32, 76, 115, 148, 170, 205 hypersurface, 7, 8, 28, 34, 65, 75, 149, 185, 188, 219 integral invariants, 76, 124–126, 207 integrally closed, 29, 30, 39, 40, 55, 56, 173, 183, 197 invariant ring, 11, 20 Jordan block, 59 Kac, 148 Kapur, 85 Karagueuzian, 8, 48, 49 Kemper, 1, 8, 10, 11, 24, 25, 47, 63, 76, 148–150, 170, 172, 173, 183, 191, 201, 205 Krull dimension, 28–33, 40, 56, 100, 195, 212, 218 Landweber, 55 Lang, 25, 27, 50 lead term, 83, 86, 87, 89, 91, 92, 94, 120, 128, 133, 137, 166, 167, 182 lead term algebra, 85 leading coefficient, 62 leading monomial, 91 linearly reductive, 41, 211, 212 Littelmann, 211 Lorenz, 172 M´ ui, 41, 46

Macdonald, 25, 27, 45 Madlener, 85 Malle, 10, 148–150 Matsumura, 25, 27 minimal free resolution, 36 minimal generating set, 35 minimal set of generators, 135 Molien, 50 Molien’s Theorem, 51–53, 61 monomial, 3, 64, 66–68, 83–86, 88, 90–92, 94, 95, 100–104, 112, 120, 123, 126–128, 133, 135–137, 145, 161, 167, 188, 205, 206, 221 Nakajima, 9, 10, 147, 185 Nakajima basis, 142, 143, 167 Nakajima group, 141, 142, 145, 147, 149, 164, 167, 185, 188, 189, 215 Nakajima’s Theorem, 144, 147 Neusel, 1 Noether, 7, 8, 11, 20, 30, 33, 41, 46, 63, 132 Noether Normalization, 30, 37 Noether number, 8, 24, 132 Noether’s degree bound, 8, 11, 46, 63 Noether’s gap, 8 Noetherian ring, 28, 31, 32, 37, 42, 47, 83, 171 orbit sum, 66–72, 91, 161 p-polynomial, 77 p-Sylow subgroup, 55, 164, 172 path height, 119–121, 123 permutation group, 40, 93, 161 permutation representation, 13, 14, 65, 66, 68, 91, 93, 94, 97, 161 Poincar´e series, 34 Pollack, 170 polynomial algebra, 7, 9, 20, 28, 34, 37, 56, 61, 149, 176, 177, 214 Popov, 211 primary invariants, 42 prime ideal, 27–30, 56, 130, 160–162, 196 pseudo-reflection, 9 reflection, 1, 8–10, 96, 97, 141, 147–149, 187

Index regular sequence, 33 regular function, 4 regular representation, 11, 53, 61, 69, 72, 74, 106, 111, 162 regular sequence, 31–33, 60, 61, 144, 171, 179, 182 relection, 150 Richman, 8, 11, 23, 115 Robbiano, 85 SAGBI basis, 45, 85–87, 91–94, 97, 115–117, 123, 182, 183, 205–207 Samuel, 27, 53 secondary invariants, 42 Selick, 49 separating invariants, 200, 201 Sereˇzkin, 150 Serre, 9, 55 Serre’s normality criterion, 55 Sezer, 132 Shank, 8, 23, 92, 115, 116, 124, 128–133, 148, 153, 170, 173, 205 Shank invariants, 128 Shephard, 9, 148, 150 Σ4 , 69, 72 Skjelbred, 170 Smith, 1, 8, 32 Stanley, 61 Steenrod, 49 Stong, 10, 55, 147 Sturmfels, 45

233

Sweedler, 85 symmetric group, 13, 44, 90, 91, 99, 101, 155 Symonds, 8, 48–50, 107 term, 36, 83, 84, 87, 93, 117, 119, 127, 129, 132, 145, 181, 182 Todd, 9, 148, 150 trace, 50, 153 transfer, 120, 135, 153–157, 159, 160, 163, 164, 179 transvection, 141, 149, 150, 167 unique factorization domain, 39, 53–55, 74, 166, 175, 182, 183, 188 upper triangular group, 141 upper triangular invariants, 46 variety, 27, 42, 161, 162 vector invariants, 11, 13, 115, 116, 173 Wagner, 150 Watanabe, 9 Wehlau, 8, 23, 92, 115, 116, 170, 173, 191, 201, 211 Wilson, 115 Woodcock, 132 Wu, 145 Zalesski˘ı, 150 Zariski, 27 Zariski topology, 25, 27